Loading...
1// SPDX-License-Identifier: GPL-2.0
2/* Copyright (c) 2018, Intel Corporation. */
3
4#include "ice.h"
5#include "ice_vf_lib_private.h"
6#include "ice_base.h"
7#include "ice_lib.h"
8#include "ice_fltr.h"
9#include "ice_dcb_lib.h"
10#include "ice_flow.h"
11#include "ice_eswitch.h"
12#include "ice_virtchnl_allowlist.h"
13#include "ice_flex_pipe.h"
14#include "ice_vf_vsi_vlan_ops.h"
15#include "ice_vlan.h"
16
17/**
18 * ice_free_vf_entries - Free all VF entries from the hash table
19 * @pf: pointer to the PF structure
20 *
21 * Iterate over the VF hash table, removing and releasing all VF entries.
22 * Called during VF teardown or as cleanup during failed VF initialization.
23 */
24static void ice_free_vf_entries(struct ice_pf *pf)
25{
26 struct ice_vfs *vfs = &pf->vfs;
27 struct hlist_node *tmp;
28 struct ice_vf *vf;
29 unsigned int bkt;
30
31 /* Remove all VFs from the hash table and release their main
32 * reference. Once all references to the VF are dropped, ice_put_vf()
33 * will call ice_release_vf which will remove the VF memory.
34 */
35 lockdep_assert_held(&vfs->table_lock);
36
37 hash_for_each_safe(vfs->table, bkt, tmp, vf, entry) {
38 hash_del_rcu(&vf->entry);
39 ice_put_vf(vf);
40 }
41}
42
43/**
44 * ice_free_vf_res - Free a VF's resources
45 * @vf: pointer to the VF info
46 */
47static void ice_free_vf_res(struct ice_vf *vf)
48{
49 struct ice_pf *pf = vf->pf;
50 int i, last_vector_idx;
51
52 /* First, disable VF's configuration API to prevent OS from
53 * accessing the VF's VSI after it's freed or invalidated.
54 */
55 clear_bit(ICE_VF_STATE_INIT, vf->vf_states);
56 ice_vf_fdir_exit(vf);
57 /* free VF control VSI */
58 if (vf->ctrl_vsi_idx != ICE_NO_VSI)
59 ice_vf_ctrl_vsi_release(vf);
60
61 /* free VSI and disconnect it from the parent uplink */
62 if (vf->lan_vsi_idx != ICE_NO_VSI) {
63 ice_vf_vsi_release(vf);
64 vf->num_mac = 0;
65 }
66
67 last_vector_idx = vf->first_vector_idx + vf->num_msix - 1;
68
69 /* clear VF MDD event information */
70 memset(&vf->mdd_tx_events, 0, sizeof(vf->mdd_tx_events));
71 memset(&vf->mdd_rx_events, 0, sizeof(vf->mdd_rx_events));
72
73 /* Disable interrupts so that VF starts in a known state */
74 for (i = vf->first_vector_idx; i <= last_vector_idx; i++) {
75 wr32(&pf->hw, GLINT_DYN_CTL(i), GLINT_DYN_CTL_CLEARPBA_M);
76 ice_flush(&pf->hw);
77 }
78 /* reset some of the state variables keeping track of the resources */
79 clear_bit(ICE_VF_STATE_MC_PROMISC, vf->vf_states);
80 clear_bit(ICE_VF_STATE_UC_PROMISC, vf->vf_states);
81}
82
83/**
84 * ice_dis_vf_mappings
85 * @vf: pointer to the VF structure
86 */
87static void ice_dis_vf_mappings(struct ice_vf *vf)
88{
89 struct ice_pf *pf = vf->pf;
90 struct ice_vsi *vsi;
91 struct device *dev;
92 int first, last, v;
93 struct ice_hw *hw;
94
95 hw = &pf->hw;
96 vsi = ice_get_vf_vsi(vf);
97 if (WARN_ON(!vsi))
98 return;
99
100 dev = ice_pf_to_dev(pf);
101 wr32(hw, VPINT_ALLOC(vf->vf_id), 0);
102 wr32(hw, VPINT_ALLOC_PCI(vf->vf_id), 0);
103
104 first = vf->first_vector_idx;
105 last = first + vf->num_msix - 1;
106 for (v = first; v <= last; v++) {
107 u32 reg;
108
109 reg = FIELD_PREP(GLINT_VECT2FUNC_IS_PF_M, 1) |
110 FIELD_PREP(GLINT_VECT2FUNC_PF_NUM_M, hw->pf_id);
111 wr32(hw, GLINT_VECT2FUNC(v), reg);
112 }
113
114 if (vsi->tx_mapping_mode == ICE_VSI_MAP_CONTIG)
115 wr32(hw, VPLAN_TX_QBASE(vf->vf_id), 0);
116 else
117 dev_err(dev, "Scattered mode for VF Tx queues is not yet implemented\n");
118
119 if (vsi->rx_mapping_mode == ICE_VSI_MAP_CONTIG)
120 wr32(hw, VPLAN_RX_QBASE(vf->vf_id), 0);
121 else
122 dev_err(dev, "Scattered mode for VF Rx queues is not yet implemented\n");
123}
124
125/**
126 * ice_sriov_free_msix_res - Reset/free any used MSIX resources
127 * @pf: pointer to the PF structure
128 *
129 * Since no MSIX entries are taken from the pf->irq_tracker then just clear
130 * the pf->sriov_base_vector.
131 *
132 * Returns 0 on success, and -EINVAL on error.
133 */
134static int ice_sriov_free_msix_res(struct ice_pf *pf)
135{
136 if (!pf)
137 return -EINVAL;
138
139 bitmap_free(pf->sriov_irq_bm);
140 pf->sriov_irq_size = 0;
141 pf->sriov_base_vector = 0;
142
143 return 0;
144}
145
146/**
147 * ice_free_vfs - Free all VFs
148 * @pf: pointer to the PF structure
149 */
150void ice_free_vfs(struct ice_pf *pf)
151{
152 struct device *dev = ice_pf_to_dev(pf);
153 struct ice_vfs *vfs = &pf->vfs;
154 struct ice_hw *hw = &pf->hw;
155 struct ice_vf *vf;
156 unsigned int bkt;
157
158 if (!ice_has_vfs(pf))
159 return;
160
161 while (test_and_set_bit(ICE_VF_DIS, pf->state))
162 usleep_range(1000, 2000);
163
164 /* Disable IOV before freeing resources. This lets any VF drivers
165 * running in the host get themselves cleaned up before we yank
166 * the carpet out from underneath their feet.
167 */
168 if (!pci_vfs_assigned(pf->pdev))
169 pci_disable_sriov(pf->pdev);
170 else
171 dev_warn(dev, "VFs are assigned - not disabling SR-IOV\n");
172
173 ice_eswitch_reserve_cp_queues(pf, -ice_get_num_vfs(pf));
174
175 mutex_lock(&vfs->table_lock);
176
177 ice_for_each_vf(pf, bkt, vf) {
178 mutex_lock(&vf->cfg_lock);
179
180 ice_eswitch_detach(pf, vf);
181 ice_dis_vf_qs(vf);
182
183 if (test_bit(ICE_VF_STATE_INIT, vf->vf_states)) {
184 /* disable VF qp mappings and set VF disable state */
185 ice_dis_vf_mappings(vf);
186 set_bit(ICE_VF_STATE_DIS, vf->vf_states);
187 ice_free_vf_res(vf);
188 }
189
190 if (!pci_vfs_assigned(pf->pdev)) {
191 u32 reg_idx, bit_idx;
192
193 reg_idx = (hw->func_caps.vf_base_id + vf->vf_id) / 32;
194 bit_idx = (hw->func_caps.vf_base_id + vf->vf_id) % 32;
195 wr32(hw, GLGEN_VFLRSTAT(reg_idx), BIT(bit_idx));
196 }
197
198 /* clear malicious info since the VF is getting released */
199 list_del(&vf->mbx_info.list_entry);
200
201 mutex_unlock(&vf->cfg_lock);
202 }
203
204 if (ice_sriov_free_msix_res(pf))
205 dev_err(dev, "Failed to free MSIX resources used by SR-IOV\n");
206
207 vfs->num_qps_per = 0;
208 ice_free_vf_entries(pf);
209
210 mutex_unlock(&vfs->table_lock);
211
212 clear_bit(ICE_VF_DIS, pf->state);
213 clear_bit(ICE_FLAG_SRIOV_ENA, pf->flags);
214}
215
216/**
217 * ice_vf_vsi_setup - Set up a VF VSI
218 * @vf: VF to setup VSI for
219 *
220 * Returns pointer to the successfully allocated VSI struct on success,
221 * otherwise returns NULL on failure.
222 */
223static struct ice_vsi *ice_vf_vsi_setup(struct ice_vf *vf)
224{
225 struct ice_vsi_cfg_params params = {};
226 struct ice_pf *pf = vf->pf;
227 struct ice_vsi *vsi;
228
229 params.type = ICE_VSI_VF;
230 params.pi = ice_vf_get_port_info(vf);
231 params.vf = vf;
232 params.flags = ICE_VSI_FLAG_INIT;
233
234 vsi = ice_vsi_setup(pf, ¶ms);
235
236 if (!vsi) {
237 dev_err(ice_pf_to_dev(pf), "Failed to create VF VSI\n");
238 ice_vf_invalidate_vsi(vf);
239 return NULL;
240 }
241
242 vf->lan_vsi_idx = vsi->idx;
243 vf->lan_vsi_num = vsi->vsi_num;
244
245 return vsi;
246}
247
248
249/**
250 * ice_ena_vf_msix_mappings - enable VF MSIX mappings in hardware
251 * @vf: VF to enable MSIX mappings for
252 *
253 * Some of the registers need to be indexed/configured using hardware global
254 * device values and other registers need 0-based values, which represent PF
255 * based values.
256 */
257static void ice_ena_vf_msix_mappings(struct ice_vf *vf)
258{
259 int device_based_first_msix, device_based_last_msix;
260 int pf_based_first_msix, pf_based_last_msix, v;
261 struct ice_pf *pf = vf->pf;
262 int device_based_vf_id;
263 struct ice_hw *hw;
264 u32 reg;
265
266 hw = &pf->hw;
267 pf_based_first_msix = vf->first_vector_idx;
268 pf_based_last_msix = (pf_based_first_msix + vf->num_msix) - 1;
269
270 device_based_first_msix = pf_based_first_msix +
271 pf->hw.func_caps.common_cap.msix_vector_first_id;
272 device_based_last_msix =
273 (device_based_first_msix + vf->num_msix) - 1;
274 device_based_vf_id = vf->vf_id + hw->func_caps.vf_base_id;
275
276 reg = FIELD_PREP(VPINT_ALLOC_FIRST_M, device_based_first_msix) |
277 FIELD_PREP(VPINT_ALLOC_LAST_M, device_based_last_msix) |
278 VPINT_ALLOC_VALID_M;
279 wr32(hw, VPINT_ALLOC(vf->vf_id), reg);
280
281 reg = FIELD_PREP(VPINT_ALLOC_PCI_FIRST_M, device_based_first_msix) |
282 FIELD_PREP(VPINT_ALLOC_PCI_LAST_M, device_based_last_msix) |
283 VPINT_ALLOC_PCI_VALID_M;
284 wr32(hw, VPINT_ALLOC_PCI(vf->vf_id), reg);
285
286 /* map the interrupts to its functions */
287 for (v = pf_based_first_msix; v <= pf_based_last_msix; v++) {
288 reg = FIELD_PREP(GLINT_VECT2FUNC_VF_NUM_M, device_based_vf_id) |
289 FIELD_PREP(GLINT_VECT2FUNC_PF_NUM_M, hw->pf_id);
290 wr32(hw, GLINT_VECT2FUNC(v), reg);
291 }
292
293 /* Map mailbox interrupt to VF MSI-X vector 0 */
294 wr32(hw, VPINT_MBX_CTL(device_based_vf_id), VPINT_MBX_CTL_CAUSE_ENA_M);
295}
296
297/**
298 * ice_ena_vf_q_mappings - enable Rx/Tx queue mappings for a VF
299 * @vf: VF to enable the mappings for
300 * @max_txq: max Tx queues allowed on the VF's VSI
301 * @max_rxq: max Rx queues allowed on the VF's VSI
302 */
303static void ice_ena_vf_q_mappings(struct ice_vf *vf, u16 max_txq, u16 max_rxq)
304{
305 struct device *dev = ice_pf_to_dev(vf->pf);
306 struct ice_vsi *vsi = ice_get_vf_vsi(vf);
307 struct ice_hw *hw = &vf->pf->hw;
308 u32 reg;
309
310 if (WARN_ON(!vsi))
311 return;
312
313 /* set regardless of mapping mode */
314 wr32(hw, VPLAN_TXQ_MAPENA(vf->vf_id), VPLAN_TXQ_MAPENA_TX_ENA_M);
315
316 /* VF Tx queues allocation */
317 if (vsi->tx_mapping_mode == ICE_VSI_MAP_CONTIG) {
318 /* set the VF PF Tx queue range
319 * VFNUMQ value should be set to (number of queues - 1). A value
320 * of 0 means 1 queue and a value of 255 means 256 queues
321 */
322 reg = FIELD_PREP(VPLAN_TX_QBASE_VFFIRSTQ_M, vsi->txq_map[0]) |
323 FIELD_PREP(VPLAN_TX_QBASE_VFNUMQ_M, max_txq - 1);
324 wr32(hw, VPLAN_TX_QBASE(vf->vf_id), reg);
325 } else {
326 dev_err(dev, "Scattered mode for VF Tx queues is not yet implemented\n");
327 }
328
329 /* set regardless of mapping mode */
330 wr32(hw, VPLAN_RXQ_MAPENA(vf->vf_id), VPLAN_RXQ_MAPENA_RX_ENA_M);
331
332 /* VF Rx queues allocation */
333 if (vsi->rx_mapping_mode == ICE_VSI_MAP_CONTIG) {
334 /* set the VF PF Rx queue range
335 * VFNUMQ value should be set to (number of queues - 1). A value
336 * of 0 means 1 queue and a value of 255 means 256 queues
337 */
338 reg = FIELD_PREP(VPLAN_RX_QBASE_VFFIRSTQ_M, vsi->rxq_map[0]) |
339 FIELD_PREP(VPLAN_RX_QBASE_VFNUMQ_M, max_rxq - 1);
340 wr32(hw, VPLAN_RX_QBASE(vf->vf_id), reg);
341 } else {
342 dev_err(dev, "Scattered mode for VF Rx queues is not yet implemented\n");
343 }
344}
345
346/**
347 * ice_ena_vf_mappings - enable VF MSIX and queue mapping
348 * @vf: pointer to the VF structure
349 */
350static void ice_ena_vf_mappings(struct ice_vf *vf)
351{
352 struct ice_vsi *vsi = ice_get_vf_vsi(vf);
353
354 if (WARN_ON(!vsi))
355 return;
356
357 ice_ena_vf_msix_mappings(vf);
358 ice_ena_vf_q_mappings(vf, vsi->alloc_txq, vsi->alloc_rxq);
359}
360
361/**
362 * ice_calc_vf_reg_idx - Calculate the VF's register index in the PF space
363 * @vf: VF to calculate the register index for
364 * @q_vector: a q_vector associated to the VF
365 */
366int ice_calc_vf_reg_idx(struct ice_vf *vf, struct ice_q_vector *q_vector)
367{
368 if (!vf || !q_vector)
369 return -EINVAL;
370
371 /* always add one to account for the OICR being the first MSIX */
372 return vf->first_vector_idx + q_vector->v_idx + 1;
373}
374
375/**
376 * ice_sriov_set_msix_res - Set any used MSIX resources
377 * @pf: pointer to PF structure
378 * @num_msix_needed: number of MSIX vectors needed for all SR-IOV VFs
379 *
380 * This function allows SR-IOV resources to be taken from the end of the PF's
381 * allowed HW MSIX vectors so that the irq_tracker will not be affected. We
382 * just set the pf->sriov_base_vector and return success.
383 *
384 * If there are not enough resources available, return an error. This should
385 * always be caught by ice_set_per_vf_res().
386 *
387 * Return 0 on success, and -EINVAL when there are not enough MSIX vectors
388 * in the PF's space available for SR-IOV.
389 */
390static int ice_sriov_set_msix_res(struct ice_pf *pf, u16 num_msix_needed)
391{
392 u16 total_vectors = pf->hw.func_caps.common_cap.num_msix_vectors;
393 int vectors_used = ice_get_max_used_msix_vector(pf);
394 int sriov_base_vector;
395
396 sriov_base_vector = total_vectors - num_msix_needed;
397
398 /* make sure we only grab irq_tracker entries from the list end and
399 * that we have enough available MSIX vectors
400 */
401 if (sriov_base_vector < vectors_used)
402 return -EINVAL;
403
404 pf->sriov_base_vector = sriov_base_vector;
405
406 return 0;
407}
408
409/**
410 * ice_set_per_vf_res - check if vectors and queues are available
411 * @pf: pointer to the PF structure
412 * @num_vfs: the number of SR-IOV VFs being configured
413 *
414 * First, determine HW interrupts from common pool. If we allocate fewer VFs, we
415 * get more vectors and can enable more queues per VF. Note that this does not
416 * grab any vectors from the SW pool already allocated. Also note, that all
417 * vector counts include one for each VF's miscellaneous interrupt vector
418 * (i.e. OICR).
419 *
420 * Minimum VFs - 2 vectors, 1 queue pair
421 * Small VFs - 5 vectors, 4 queue pairs
422 * Medium VFs - 17 vectors, 16 queue pairs
423 *
424 * Second, determine number of queue pairs per VF by starting with a pre-defined
425 * maximum each VF supports. If this is not possible, then we adjust based on
426 * queue pairs available on the device.
427 *
428 * Lastly, set queue and MSI-X VF variables tracked by the PF so it can be used
429 * by each VF during VF initialization and reset.
430 */
431static int ice_set_per_vf_res(struct ice_pf *pf, u16 num_vfs)
432{
433 int vectors_used = ice_get_max_used_msix_vector(pf);
434 u16 num_msix_per_vf, num_txq, num_rxq, avail_qs;
435 int msix_avail_per_vf, msix_avail_for_sriov;
436 struct device *dev = ice_pf_to_dev(pf);
437 int err;
438
439 lockdep_assert_held(&pf->vfs.table_lock);
440
441 if (!num_vfs)
442 return -EINVAL;
443
444 /* determine MSI-X resources per VF */
445 msix_avail_for_sriov = pf->hw.func_caps.common_cap.num_msix_vectors -
446 vectors_used;
447 msix_avail_per_vf = msix_avail_for_sriov / num_vfs;
448 if (msix_avail_per_vf >= ICE_NUM_VF_MSIX_MED) {
449 num_msix_per_vf = ICE_NUM_VF_MSIX_MED;
450 } else if (msix_avail_per_vf >= ICE_NUM_VF_MSIX_SMALL) {
451 num_msix_per_vf = ICE_NUM_VF_MSIX_SMALL;
452 } else if (msix_avail_per_vf >= ICE_NUM_VF_MSIX_MULTIQ_MIN) {
453 num_msix_per_vf = ICE_NUM_VF_MSIX_MULTIQ_MIN;
454 } else if (msix_avail_per_vf >= ICE_MIN_INTR_PER_VF) {
455 num_msix_per_vf = ICE_MIN_INTR_PER_VF;
456 } else {
457 dev_err(dev, "Only %d MSI-X interrupts available for SR-IOV. Not enough to support minimum of %d MSI-X interrupts per VF for %d VFs\n",
458 msix_avail_for_sriov, ICE_MIN_INTR_PER_VF,
459 num_vfs);
460 return -ENOSPC;
461 }
462
463 num_txq = min_t(u16, num_msix_per_vf - ICE_NONQ_VECS_VF,
464 ICE_MAX_RSS_QS_PER_VF);
465 avail_qs = ice_get_avail_txq_count(pf) / num_vfs;
466 if (!avail_qs)
467 num_txq = 0;
468 else if (num_txq > avail_qs)
469 num_txq = rounddown_pow_of_two(avail_qs);
470
471 num_rxq = min_t(u16, num_msix_per_vf - ICE_NONQ_VECS_VF,
472 ICE_MAX_RSS_QS_PER_VF);
473 avail_qs = ice_get_avail_rxq_count(pf) / num_vfs;
474 if (!avail_qs)
475 num_rxq = 0;
476 else if (num_rxq > avail_qs)
477 num_rxq = rounddown_pow_of_two(avail_qs);
478
479 if (num_txq < ICE_MIN_QS_PER_VF || num_rxq < ICE_MIN_QS_PER_VF) {
480 dev_err(dev, "Not enough queues to support minimum of %d queue pairs per VF for %d VFs\n",
481 ICE_MIN_QS_PER_VF, num_vfs);
482 return -ENOSPC;
483 }
484
485 err = ice_sriov_set_msix_res(pf, num_msix_per_vf * num_vfs);
486 if (err) {
487 dev_err(dev, "Unable to set MSI-X resources for %d VFs, err %d\n",
488 num_vfs, err);
489 return err;
490 }
491
492 /* only allow equal Tx/Rx queue count (i.e. queue pairs) */
493 pf->vfs.num_qps_per = min_t(int, num_txq, num_rxq);
494 pf->vfs.num_msix_per = num_msix_per_vf;
495 dev_info(dev, "Enabling %d VFs with %d vectors and %d queues per VF\n",
496 num_vfs, pf->vfs.num_msix_per, pf->vfs.num_qps_per);
497
498 return 0;
499}
500
501/**
502 * ice_sriov_get_irqs - get irqs for SR-IOV usacase
503 * @pf: pointer to PF structure
504 * @needed: number of irqs to get
505 *
506 * This returns the first MSI-X vector index in PF space that is used by this
507 * VF. This index is used when accessing PF relative registers such as
508 * GLINT_VECT2FUNC and GLINT_DYN_CTL.
509 * This will always be the OICR index in the AVF driver so any functionality
510 * using vf->first_vector_idx for queue configuration_id: id of VF which will
511 * use this irqs
512 *
513 * Only SRIOV specific vectors are tracked in sriov_irq_bm. SRIOV vectors are
514 * allocated from the end of global irq index. First bit in sriov_irq_bm means
515 * last irq index etc. It simplifies extension of SRIOV vectors.
516 * They will be always located from sriov_base_vector to the last irq
517 * index. While increasing/decreasing sriov_base_vector can be moved.
518 */
519static int ice_sriov_get_irqs(struct ice_pf *pf, u16 needed)
520{
521 int res = bitmap_find_next_zero_area(pf->sriov_irq_bm,
522 pf->sriov_irq_size, 0, needed, 0);
523 /* conversion from number in bitmap to global irq index */
524 int index = pf->sriov_irq_size - res - needed;
525
526 if (res >= pf->sriov_irq_size || index < pf->sriov_base_vector)
527 return -ENOENT;
528
529 bitmap_set(pf->sriov_irq_bm, res, needed);
530 return index;
531}
532
533/**
534 * ice_sriov_free_irqs - free irqs used by the VF
535 * @pf: pointer to PF structure
536 * @vf: pointer to VF structure
537 */
538static void ice_sriov_free_irqs(struct ice_pf *pf, struct ice_vf *vf)
539{
540 /* Move back from first vector index to first index in bitmap */
541 int bm_i = pf->sriov_irq_size - vf->first_vector_idx - vf->num_msix;
542
543 bitmap_clear(pf->sriov_irq_bm, bm_i, vf->num_msix);
544 vf->first_vector_idx = 0;
545}
546
547/**
548 * ice_init_vf_vsi_res - initialize/setup VF VSI resources
549 * @vf: VF to initialize/setup the VSI for
550 *
551 * This function creates a VSI for the VF, adds a VLAN 0 filter, and sets up the
552 * VF VSI's broadcast filter and is only used during initial VF creation.
553 */
554static int ice_init_vf_vsi_res(struct ice_vf *vf)
555{
556 struct ice_pf *pf = vf->pf;
557 struct ice_vsi *vsi;
558 int err;
559
560 vf->first_vector_idx = ice_sriov_get_irqs(pf, vf->num_msix);
561 if (vf->first_vector_idx < 0)
562 return -ENOMEM;
563
564 vsi = ice_vf_vsi_setup(vf);
565 if (!vsi)
566 return -ENOMEM;
567
568 err = ice_vf_init_host_cfg(vf, vsi);
569 if (err)
570 goto release_vsi;
571
572 return 0;
573
574release_vsi:
575 ice_vf_vsi_release(vf);
576 return err;
577}
578
579/**
580 * ice_start_vfs - start VFs so they are ready to be used by SR-IOV
581 * @pf: PF the VFs are associated with
582 */
583static int ice_start_vfs(struct ice_pf *pf)
584{
585 struct ice_hw *hw = &pf->hw;
586 unsigned int bkt, it_cnt;
587 struct ice_vf *vf;
588 int retval;
589
590 lockdep_assert_held(&pf->vfs.table_lock);
591
592 it_cnt = 0;
593 ice_for_each_vf(pf, bkt, vf) {
594 vf->vf_ops->clear_reset_trigger(vf);
595
596 retval = ice_init_vf_vsi_res(vf);
597 if (retval) {
598 dev_err(ice_pf_to_dev(pf), "Failed to initialize VSI resources for VF %d, error %d\n",
599 vf->vf_id, retval);
600 goto teardown;
601 }
602
603 retval = ice_eswitch_attach(pf, vf);
604 if (retval) {
605 dev_err(ice_pf_to_dev(pf), "Failed to attach VF %d to eswitch, error %d",
606 vf->vf_id, retval);
607 ice_vf_vsi_release(vf);
608 goto teardown;
609 }
610
611 set_bit(ICE_VF_STATE_INIT, vf->vf_states);
612 ice_ena_vf_mappings(vf);
613 wr32(hw, VFGEN_RSTAT(vf->vf_id), VIRTCHNL_VFR_VFACTIVE);
614 it_cnt++;
615 }
616
617 ice_flush(hw);
618 return 0;
619
620teardown:
621 ice_for_each_vf(pf, bkt, vf) {
622 if (it_cnt == 0)
623 break;
624
625 ice_dis_vf_mappings(vf);
626 ice_vf_vsi_release(vf);
627 it_cnt--;
628 }
629
630 return retval;
631}
632
633/**
634 * ice_sriov_free_vf - Free VF memory after all references are dropped
635 * @vf: pointer to VF to free
636 *
637 * Called by ice_put_vf through ice_release_vf once the last reference to a VF
638 * structure has been dropped.
639 */
640static void ice_sriov_free_vf(struct ice_vf *vf)
641{
642 mutex_destroy(&vf->cfg_lock);
643
644 kfree_rcu(vf, rcu);
645}
646
647/**
648 * ice_sriov_clear_reset_state - clears VF Reset status register
649 * @vf: the vf to configure
650 */
651static void ice_sriov_clear_reset_state(struct ice_vf *vf)
652{
653 struct ice_hw *hw = &vf->pf->hw;
654
655 /* Clear the reset status register so that VF immediately sees that
656 * the device is resetting, even if hardware hasn't yet gotten around
657 * to clearing VFGEN_RSTAT for us.
658 */
659 wr32(hw, VFGEN_RSTAT(vf->vf_id), VIRTCHNL_VFR_INPROGRESS);
660}
661
662/**
663 * ice_sriov_clear_mbx_register - clears SRIOV VF's mailbox registers
664 * @vf: the vf to configure
665 */
666static void ice_sriov_clear_mbx_register(struct ice_vf *vf)
667{
668 struct ice_pf *pf = vf->pf;
669
670 wr32(&pf->hw, VF_MBX_ARQLEN(vf->vf_id), 0);
671 wr32(&pf->hw, VF_MBX_ATQLEN(vf->vf_id), 0);
672}
673
674/**
675 * ice_sriov_trigger_reset_register - trigger VF reset for SRIOV VF
676 * @vf: pointer to VF structure
677 * @is_vflr: true if reset occurred due to VFLR
678 *
679 * Trigger and cleanup after a VF reset for a SR-IOV VF.
680 */
681static void ice_sriov_trigger_reset_register(struct ice_vf *vf, bool is_vflr)
682{
683 struct ice_pf *pf = vf->pf;
684 u32 reg, reg_idx, bit_idx;
685 unsigned int vf_abs_id, i;
686 struct device *dev;
687 struct ice_hw *hw;
688
689 dev = ice_pf_to_dev(pf);
690 hw = &pf->hw;
691 vf_abs_id = vf->vf_id + hw->func_caps.vf_base_id;
692
693 /* In the case of a VFLR, HW has already reset the VF and we just need
694 * to clean up. Otherwise we must first trigger the reset using the
695 * VFRTRIG register.
696 */
697 if (!is_vflr) {
698 reg = rd32(hw, VPGEN_VFRTRIG(vf->vf_id));
699 reg |= VPGEN_VFRTRIG_VFSWR_M;
700 wr32(hw, VPGEN_VFRTRIG(vf->vf_id), reg);
701 }
702
703 /* clear the VFLR bit in GLGEN_VFLRSTAT */
704 reg_idx = (vf_abs_id) / 32;
705 bit_idx = (vf_abs_id) % 32;
706 wr32(hw, GLGEN_VFLRSTAT(reg_idx), BIT(bit_idx));
707 ice_flush(hw);
708
709 wr32(hw, PF_PCI_CIAA,
710 VF_DEVICE_STATUS | (vf_abs_id << PF_PCI_CIAA_VF_NUM_S));
711 for (i = 0; i < ICE_PCI_CIAD_WAIT_COUNT; i++) {
712 reg = rd32(hw, PF_PCI_CIAD);
713 /* no transactions pending so stop polling */
714 if ((reg & VF_TRANS_PENDING_M) == 0)
715 break;
716
717 dev_err(dev, "VF %u PCI transactions stuck\n", vf->vf_id);
718 udelay(ICE_PCI_CIAD_WAIT_DELAY_US);
719 }
720}
721
722/**
723 * ice_sriov_poll_reset_status - poll SRIOV VF reset status
724 * @vf: pointer to VF structure
725 *
726 * Returns true when reset is successful, else returns false
727 */
728static bool ice_sriov_poll_reset_status(struct ice_vf *vf)
729{
730 struct ice_pf *pf = vf->pf;
731 unsigned int i;
732 u32 reg;
733
734 for (i = 0; i < 10; i++) {
735 /* VF reset requires driver to first reset the VF and then
736 * poll the status register to make sure that the reset
737 * completed successfully.
738 */
739 reg = rd32(&pf->hw, VPGEN_VFRSTAT(vf->vf_id));
740 if (reg & VPGEN_VFRSTAT_VFRD_M)
741 return true;
742
743 /* only sleep if the reset is not done */
744 usleep_range(10, 20);
745 }
746 return false;
747}
748
749/**
750 * ice_sriov_clear_reset_trigger - enable VF to access hardware
751 * @vf: VF to enabled hardware access for
752 */
753static void ice_sriov_clear_reset_trigger(struct ice_vf *vf)
754{
755 struct ice_hw *hw = &vf->pf->hw;
756 u32 reg;
757
758 reg = rd32(hw, VPGEN_VFRTRIG(vf->vf_id));
759 reg &= ~VPGEN_VFRTRIG_VFSWR_M;
760 wr32(hw, VPGEN_VFRTRIG(vf->vf_id), reg);
761 ice_flush(hw);
762}
763
764/**
765 * ice_sriov_post_vsi_rebuild - tasks to do after the VF's VSI have been rebuilt
766 * @vf: VF to perform tasks on
767 */
768static void ice_sriov_post_vsi_rebuild(struct ice_vf *vf)
769{
770 ice_ena_vf_mappings(vf);
771 wr32(&vf->pf->hw, VFGEN_RSTAT(vf->vf_id), VIRTCHNL_VFR_VFACTIVE);
772}
773
774static const struct ice_vf_ops ice_sriov_vf_ops = {
775 .reset_type = ICE_VF_RESET,
776 .free = ice_sriov_free_vf,
777 .clear_reset_state = ice_sriov_clear_reset_state,
778 .clear_mbx_register = ice_sriov_clear_mbx_register,
779 .trigger_reset_register = ice_sriov_trigger_reset_register,
780 .poll_reset_status = ice_sriov_poll_reset_status,
781 .clear_reset_trigger = ice_sriov_clear_reset_trigger,
782 .irq_close = NULL,
783 .post_vsi_rebuild = ice_sriov_post_vsi_rebuild,
784};
785
786/**
787 * ice_create_vf_entries - Allocate and insert VF entries
788 * @pf: pointer to the PF structure
789 * @num_vfs: the number of VFs to allocate
790 *
791 * Allocate new VF entries and insert them into the hash table. Set some
792 * basic default fields for initializing the new VFs.
793 *
794 * After this function exits, the hash table will have num_vfs entries
795 * inserted.
796 *
797 * Returns 0 on success or an integer error code on failure.
798 */
799static int ice_create_vf_entries(struct ice_pf *pf, u16 num_vfs)
800{
801 struct pci_dev *pdev = pf->pdev;
802 struct ice_vfs *vfs = &pf->vfs;
803 struct pci_dev *vfdev = NULL;
804 struct ice_vf *vf;
805 u16 vf_pdev_id;
806 int err, pos;
807
808 lockdep_assert_held(&vfs->table_lock);
809
810 pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
811 pci_read_config_word(pdev, pos + PCI_SRIOV_VF_DID, &vf_pdev_id);
812
813 for (u16 vf_id = 0; vf_id < num_vfs; vf_id++) {
814 vf = kzalloc(sizeof(*vf), GFP_KERNEL);
815 if (!vf) {
816 err = -ENOMEM;
817 goto err_free_entries;
818 }
819 kref_init(&vf->refcnt);
820
821 vf->pf = pf;
822 vf->vf_id = vf_id;
823
824 /* set sriov vf ops for VFs created during SRIOV flow */
825 vf->vf_ops = &ice_sriov_vf_ops;
826
827 ice_initialize_vf_entry(vf);
828
829 do {
830 vfdev = pci_get_device(pdev->vendor, vf_pdev_id, vfdev);
831 } while (vfdev && vfdev->physfn != pdev);
832 vf->vfdev = vfdev;
833 vf->vf_sw_id = pf->first_sw;
834
835 pci_dev_get(vfdev);
836
837 /* set default number of MSI-X */
838 vf->num_msix = pf->vfs.num_msix_per;
839 vf->num_vf_qs = pf->vfs.num_qps_per;
840 ice_vc_set_default_allowlist(vf);
841
842 hash_add_rcu(vfs->table, &vf->entry, vf_id);
843 }
844
845 /* Decrement of refcount done by pci_get_device() inside the loop does
846 * not touch the last iteration's vfdev, so it has to be done manually
847 * to balance pci_dev_get() added within the loop.
848 */
849 pci_dev_put(vfdev);
850
851 return 0;
852
853err_free_entries:
854 ice_free_vf_entries(pf);
855 return err;
856}
857
858/**
859 * ice_ena_vfs - enable VFs so they are ready to be used
860 * @pf: pointer to the PF structure
861 * @num_vfs: number of VFs to enable
862 */
863static int ice_ena_vfs(struct ice_pf *pf, u16 num_vfs)
864{
865 int total_vectors = pf->hw.func_caps.common_cap.num_msix_vectors;
866 struct device *dev = ice_pf_to_dev(pf);
867 struct ice_hw *hw = &pf->hw;
868 int ret;
869
870 pf->sriov_irq_bm = bitmap_zalloc(total_vectors, GFP_KERNEL);
871 if (!pf->sriov_irq_bm)
872 return -ENOMEM;
873 pf->sriov_irq_size = total_vectors;
874
875 /* Disable global interrupt 0 so we don't try to handle the VFLR. */
876 wr32(hw, GLINT_DYN_CTL(pf->oicr_irq.index),
877 ICE_ITR_NONE << GLINT_DYN_CTL_ITR_INDX_S);
878 set_bit(ICE_OICR_INTR_DIS, pf->state);
879 ice_flush(hw);
880
881 ret = pci_enable_sriov(pf->pdev, num_vfs);
882 if (ret)
883 goto err_unroll_intr;
884
885 mutex_lock(&pf->vfs.table_lock);
886
887 ret = ice_set_per_vf_res(pf, num_vfs);
888 if (ret) {
889 dev_err(dev, "Not enough resources for %d VFs, err %d. Try with fewer number of VFs\n",
890 num_vfs, ret);
891 goto err_unroll_sriov;
892 }
893
894 ret = ice_create_vf_entries(pf, num_vfs);
895 if (ret) {
896 dev_err(dev, "Failed to allocate VF entries for %d VFs\n",
897 num_vfs);
898 goto err_unroll_sriov;
899 }
900
901 ice_eswitch_reserve_cp_queues(pf, num_vfs);
902 ret = ice_start_vfs(pf);
903 if (ret) {
904 dev_err(dev, "Failed to start %d VFs, err %d\n", num_vfs, ret);
905 ret = -EAGAIN;
906 goto err_unroll_vf_entries;
907 }
908
909 clear_bit(ICE_VF_DIS, pf->state);
910
911 /* rearm global interrupts */
912 if (test_and_clear_bit(ICE_OICR_INTR_DIS, pf->state))
913 ice_irq_dynamic_ena(hw, NULL, NULL);
914
915 mutex_unlock(&pf->vfs.table_lock);
916
917 return 0;
918
919err_unroll_vf_entries:
920 ice_free_vf_entries(pf);
921err_unroll_sriov:
922 mutex_unlock(&pf->vfs.table_lock);
923 pci_disable_sriov(pf->pdev);
924err_unroll_intr:
925 /* rearm interrupts here */
926 ice_irq_dynamic_ena(hw, NULL, NULL);
927 clear_bit(ICE_OICR_INTR_DIS, pf->state);
928 bitmap_free(pf->sriov_irq_bm);
929 return ret;
930}
931
932/**
933 * ice_pci_sriov_ena - Enable or change number of VFs
934 * @pf: pointer to the PF structure
935 * @num_vfs: number of VFs to allocate
936 *
937 * Returns 0 on success and negative on failure
938 */
939static int ice_pci_sriov_ena(struct ice_pf *pf, int num_vfs)
940{
941 struct device *dev = ice_pf_to_dev(pf);
942 int err;
943
944 if (!num_vfs) {
945 ice_free_vfs(pf);
946 return 0;
947 }
948
949 if (num_vfs > pf->vfs.num_supported) {
950 dev_err(dev, "Can't enable %d VFs, max VFs supported is %d\n",
951 num_vfs, pf->vfs.num_supported);
952 return -EOPNOTSUPP;
953 }
954
955 dev_info(dev, "Enabling %d VFs\n", num_vfs);
956 err = ice_ena_vfs(pf, num_vfs);
957 if (err) {
958 dev_err(dev, "Failed to enable SR-IOV: %d\n", err);
959 return err;
960 }
961
962 set_bit(ICE_FLAG_SRIOV_ENA, pf->flags);
963 return 0;
964}
965
966/**
967 * ice_check_sriov_allowed - check if SR-IOV is allowed based on various checks
968 * @pf: PF to enabled SR-IOV on
969 */
970static int ice_check_sriov_allowed(struct ice_pf *pf)
971{
972 struct device *dev = ice_pf_to_dev(pf);
973
974 if (!test_bit(ICE_FLAG_SRIOV_CAPABLE, pf->flags)) {
975 dev_err(dev, "This device is not capable of SR-IOV\n");
976 return -EOPNOTSUPP;
977 }
978
979 if (ice_is_safe_mode(pf)) {
980 dev_err(dev, "SR-IOV cannot be configured - Device is in Safe Mode\n");
981 return -EOPNOTSUPP;
982 }
983
984 if (!ice_pf_state_is_nominal(pf)) {
985 dev_err(dev, "Cannot enable SR-IOV, device not ready\n");
986 return -EBUSY;
987 }
988
989 return 0;
990}
991
992/**
993 * ice_sriov_get_vf_total_msix - return number of MSI-X used by VFs
994 * @pdev: pointer to pci_dev struct
995 *
996 * The function is called via sysfs ops
997 */
998u32 ice_sriov_get_vf_total_msix(struct pci_dev *pdev)
999{
1000 struct ice_pf *pf = pci_get_drvdata(pdev);
1001
1002 return pf->sriov_irq_size - ice_get_max_used_msix_vector(pf);
1003}
1004
1005static int ice_sriov_move_base_vector(struct ice_pf *pf, int move)
1006{
1007 if (pf->sriov_base_vector - move < ice_get_max_used_msix_vector(pf))
1008 return -ENOMEM;
1009
1010 pf->sriov_base_vector -= move;
1011 return 0;
1012}
1013
1014static void ice_sriov_remap_vectors(struct ice_pf *pf, u16 restricted_id)
1015{
1016 u16 vf_ids[ICE_MAX_SRIOV_VFS];
1017 struct ice_vf *tmp_vf;
1018 int to_remap = 0, bkt;
1019
1020 /* For better irqs usage try to remap irqs of VFs
1021 * that aren't running yet
1022 */
1023 ice_for_each_vf(pf, bkt, tmp_vf) {
1024 /* skip VF which is changing the number of MSI-X */
1025 if (restricted_id == tmp_vf->vf_id ||
1026 test_bit(ICE_VF_STATE_ACTIVE, tmp_vf->vf_states))
1027 continue;
1028
1029 ice_dis_vf_mappings(tmp_vf);
1030 ice_sriov_free_irqs(pf, tmp_vf);
1031
1032 vf_ids[to_remap] = tmp_vf->vf_id;
1033 to_remap += 1;
1034 }
1035
1036 for (int i = 0; i < to_remap; i++) {
1037 tmp_vf = ice_get_vf_by_id(pf, vf_ids[i]);
1038 if (!tmp_vf)
1039 continue;
1040
1041 tmp_vf->first_vector_idx =
1042 ice_sriov_get_irqs(pf, tmp_vf->num_msix);
1043 /* there is no need to rebuild VSI as we are only changing the
1044 * vector indexes not amount of MSI-X or queues
1045 */
1046 ice_ena_vf_mappings(tmp_vf);
1047 ice_put_vf(tmp_vf);
1048 }
1049}
1050
1051/**
1052 * ice_sriov_set_msix_vec_count
1053 * @vf_dev: pointer to pci_dev struct of VF device
1054 * @msix_vec_count: new value for MSI-X amount on this VF
1055 *
1056 * Set requested MSI-X, queues and registers for @vf_dev.
1057 *
1058 * First do some sanity checks like if there are any VFs, if the new value
1059 * is correct etc. Then disable old mapping (MSI-X and queues registers), change
1060 * MSI-X and queues, rebuild VSI and enable new mapping.
1061 *
1062 * If it is possible (driver not binded to VF) try to remap also other VFs to
1063 * linearize irqs register usage.
1064 */
1065int ice_sriov_set_msix_vec_count(struct pci_dev *vf_dev, int msix_vec_count)
1066{
1067 struct pci_dev *pdev = pci_physfn(vf_dev);
1068 struct ice_pf *pf = pci_get_drvdata(pdev);
1069 u16 prev_msix, prev_queues, queues;
1070 bool needs_rebuild = false;
1071 struct ice_vsi *vsi;
1072 struct ice_vf *vf;
1073 int id;
1074
1075 if (!ice_get_num_vfs(pf))
1076 return -ENOENT;
1077
1078 if (!msix_vec_count)
1079 return 0;
1080
1081 queues = msix_vec_count;
1082 /* add 1 MSI-X for OICR */
1083 msix_vec_count += 1;
1084
1085 if (queues > min(ice_get_avail_txq_count(pf),
1086 ice_get_avail_rxq_count(pf)))
1087 return -EINVAL;
1088
1089 if (msix_vec_count < ICE_MIN_INTR_PER_VF)
1090 return -EINVAL;
1091
1092 /* Transition of PCI VF function number to function_id */
1093 for (id = 0; id < pci_num_vf(pdev); id++) {
1094 if (vf_dev->devfn == pci_iov_virtfn_devfn(pdev, id))
1095 break;
1096 }
1097
1098 if (id == pci_num_vf(pdev))
1099 return -ENOENT;
1100
1101 vf = ice_get_vf_by_id(pf, id);
1102
1103 if (!vf)
1104 return -ENOENT;
1105
1106 vsi = ice_get_vf_vsi(vf);
1107 if (!vsi)
1108 return -ENOENT;
1109
1110 prev_msix = vf->num_msix;
1111 prev_queues = vf->num_vf_qs;
1112
1113 if (ice_sriov_move_base_vector(pf, msix_vec_count - prev_msix)) {
1114 ice_put_vf(vf);
1115 return -ENOSPC;
1116 }
1117
1118 ice_dis_vf_mappings(vf);
1119 ice_sriov_free_irqs(pf, vf);
1120
1121 /* Remap all VFs beside the one is now configured */
1122 ice_sriov_remap_vectors(pf, vf->vf_id);
1123
1124 vf->num_msix = msix_vec_count;
1125 vf->num_vf_qs = queues;
1126 vf->first_vector_idx = ice_sriov_get_irqs(pf, vf->num_msix);
1127 if (vf->first_vector_idx < 0)
1128 goto unroll;
1129
1130 if (ice_vf_reconfig_vsi(vf) || ice_vf_init_host_cfg(vf, vsi)) {
1131 /* Try to rebuild with previous values */
1132 needs_rebuild = true;
1133 goto unroll;
1134 }
1135
1136 dev_info(ice_pf_to_dev(pf),
1137 "Changing VF %d resources to %d vectors and %d queues\n",
1138 vf->vf_id, vf->num_msix, vf->num_vf_qs);
1139
1140 ice_ena_vf_mappings(vf);
1141 ice_put_vf(vf);
1142
1143 return 0;
1144
1145unroll:
1146 dev_info(ice_pf_to_dev(pf),
1147 "Can't set %d vectors on VF %d, falling back to %d\n",
1148 vf->num_msix, vf->vf_id, prev_msix);
1149
1150 vf->num_msix = prev_msix;
1151 vf->num_vf_qs = prev_queues;
1152 vf->first_vector_idx = ice_sriov_get_irqs(pf, vf->num_msix);
1153 if (vf->first_vector_idx < 0)
1154 return -EINVAL;
1155
1156 if (needs_rebuild) {
1157 ice_vf_reconfig_vsi(vf);
1158 ice_vf_init_host_cfg(vf, vsi);
1159 }
1160
1161 ice_ena_vf_mappings(vf);
1162 ice_put_vf(vf);
1163
1164 return -EINVAL;
1165}
1166
1167/**
1168 * ice_sriov_configure - Enable or change number of VFs via sysfs
1169 * @pdev: pointer to a pci_dev structure
1170 * @num_vfs: number of VFs to allocate or 0 to free VFs
1171 *
1172 * This function is called when the user updates the number of VFs in sysfs. On
1173 * success return whatever num_vfs was set to by the caller. Return negative on
1174 * failure.
1175 */
1176int ice_sriov_configure(struct pci_dev *pdev, int num_vfs)
1177{
1178 struct ice_pf *pf = pci_get_drvdata(pdev);
1179 struct device *dev = ice_pf_to_dev(pf);
1180 int err;
1181
1182 err = ice_check_sriov_allowed(pf);
1183 if (err)
1184 return err;
1185
1186 if (!num_vfs) {
1187 if (!pci_vfs_assigned(pdev)) {
1188 ice_free_vfs(pf);
1189 return 0;
1190 }
1191
1192 dev_err(dev, "can't free VFs because some are assigned to VMs.\n");
1193 return -EBUSY;
1194 }
1195
1196 err = ice_pci_sriov_ena(pf, num_vfs);
1197 if (err)
1198 return err;
1199
1200 return num_vfs;
1201}
1202
1203/**
1204 * ice_process_vflr_event - Free VF resources via IRQ calls
1205 * @pf: pointer to the PF structure
1206 *
1207 * called from the VFLR IRQ handler to
1208 * free up VF resources and state variables
1209 */
1210void ice_process_vflr_event(struct ice_pf *pf)
1211{
1212 struct ice_hw *hw = &pf->hw;
1213 struct ice_vf *vf;
1214 unsigned int bkt;
1215 u32 reg;
1216
1217 if (!test_and_clear_bit(ICE_VFLR_EVENT_PENDING, pf->state) ||
1218 !ice_has_vfs(pf))
1219 return;
1220
1221 mutex_lock(&pf->vfs.table_lock);
1222 ice_for_each_vf(pf, bkt, vf) {
1223 u32 reg_idx, bit_idx;
1224
1225 reg_idx = (hw->func_caps.vf_base_id + vf->vf_id) / 32;
1226 bit_idx = (hw->func_caps.vf_base_id + vf->vf_id) % 32;
1227 /* read GLGEN_VFLRSTAT register to find out the flr VFs */
1228 reg = rd32(hw, GLGEN_VFLRSTAT(reg_idx));
1229 if (reg & BIT(bit_idx))
1230 /* GLGEN_VFLRSTAT bit will be cleared in ice_reset_vf */
1231 ice_reset_vf(vf, ICE_VF_RESET_VFLR | ICE_VF_RESET_LOCK);
1232 }
1233 mutex_unlock(&pf->vfs.table_lock);
1234}
1235
1236/**
1237 * ice_get_vf_from_pfq - get the VF who owns the PF space queue passed in
1238 * @pf: PF used to index all VFs
1239 * @pfq: queue index relative to the PF's function space
1240 *
1241 * If no VF is found who owns the pfq then return NULL, otherwise return a
1242 * pointer to the VF who owns the pfq
1243 *
1244 * If this function returns non-NULL, it acquires a reference count of the VF
1245 * structure. The caller is responsible for calling ice_put_vf() to drop this
1246 * reference.
1247 */
1248static struct ice_vf *ice_get_vf_from_pfq(struct ice_pf *pf, u16 pfq)
1249{
1250 struct ice_vf *vf;
1251 unsigned int bkt;
1252
1253 rcu_read_lock();
1254 ice_for_each_vf_rcu(pf, bkt, vf) {
1255 struct ice_vsi *vsi;
1256 u16 rxq_idx;
1257
1258 vsi = ice_get_vf_vsi(vf);
1259 if (!vsi)
1260 continue;
1261
1262 ice_for_each_rxq(vsi, rxq_idx)
1263 if (vsi->rxq_map[rxq_idx] == pfq) {
1264 struct ice_vf *found;
1265
1266 if (kref_get_unless_zero(&vf->refcnt))
1267 found = vf;
1268 else
1269 found = NULL;
1270 rcu_read_unlock();
1271 return found;
1272 }
1273 }
1274 rcu_read_unlock();
1275
1276 return NULL;
1277}
1278
1279/**
1280 * ice_globalq_to_pfq - convert from global queue index to PF space queue index
1281 * @pf: PF used for conversion
1282 * @globalq: global queue index used to convert to PF space queue index
1283 */
1284static u32 ice_globalq_to_pfq(struct ice_pf *pf, u32 globalq)
1285{
1286 return globalq - pf->hw.func_caps.common_cap.rxq_first_id;
1287}
1288
1289/**
1290 * ice_vf_lan_overflow_event - handle LAN overflow event for a VF
1291 * @pf: PF that the LAN overflow event happened on
1292 * @event: structure holding the event information for the LAN overflow event
1293 *
1294 * Determine if the LAN overflow event was caused by a VF queue. If it was not
1295 * caused by a VF, do nothing. If a VF caused this LAN overflow event trigger a
1296 * reset on the offending VF.
1297 */
1298void
1299ice_vf_lan_overflow_event(struct ice_pf *pf, struct ice_rq_event_info *event)
1300{
1301 u32 gldcb_rtctq, queue;
1302 struct ice_vf *vf;
1303
1304 gldcb_rtctq = le32_to_cpu(event->desc.params.lan_overflow.prtdcb_ruptq);
1305 dev_dbg(ice_pf_to_dev(pf), "GLDCB_RTCTQ: 0x%08x\n", gldcb_rtctq);
1306
1307 /* event returns device global Rx queue number */
1308 queue = FIELD_GET(GLDCB_RTCTQ_RXQNUM_M, gldcb_rtctq);
1309
1310 vf = ice_get_vf_from_pfq(pf, ice_globalq_to_pfq(pf, queue));
1311 if (!vf)
1312 return;
1313
1314 ice_reset_vf(vf, ICE_VF_RESET_NOTIFY | ICE_VF_RESET_LOCK);
1315 ice_put_vf(vf);
1316}
1317
1318/**
1319 * ice_set_vf_spoofchk
1320 * @netdev: network interface device structure
1321 * @vf_id: VF identifier
1322 * @ena: flag to enable or disable feature
1323 *
1324 * Enable or disable VF spoof checking
1325 */
1326int ice_set_vf_spoofchk(struct net_device *netdev, int vf_id, bool ena)
1327{
1328 struct ice_netdev_priv *np = netdev_priv(netdev);
1329 struct ice_pf *pf = np->vsi->back;
1330 struct ice_vsi *vf_vsi;
1331 struct device *dev;
1332 struct ice_vf *vf;
1333 int ret;
1334
1335 dev = ice_pf_to_dev(pf);
1336
1337 vf = ice_get_vf_by_id(pf, vf_id);
1338 if (!vf)
1339 return -EINVAL;
1340
1341 ret = ice_check_vf_ready_for_cfg(vf);
1342 if (ret)
1343 goto out_put_vf;
1344
1345 vf_vsi = ice_get_vf_vsi(vf);
1346 if (!vf_vsi) {
1347 netdev_err(netdev, "VSI %d for VF %d is null\n",
1348 vf->lan_vsi_idx, vf->vf_id);
1349 ret = -EINVAL;
1350 goto out_put_vf;
1351 }
1352
1353 if (vf_vsi->type != ICE_VSI_VF) {
1354 netdev_err(netdev, "Type %d of VSI %d for VF %d is no ICE_VSI_VF\n",
1355 vf_vsi->type, vf_vsi->vsi_num, vf->vf_id);
1356 ret = -ENODEV;
1357 goto out_put_vf;
1358 }
1359
1360 if (ena == vf->spoofchk) {
1361 dev_dbg(dev, "VF spoofchk already %s\n", ena ? "ON" : "OFF");
1362 ret = 0;
1363 goto out_put_vf;
1364 }
1365
1366 ret = ice_vsi_apply_spoofchk(vf_vsi, ena);
1367 if (ret)
1368 dev_err(dev, "Failed to set spoofchk %s for VF %d VSI %d\n error %d\n",
1369 ena ? "ON" : "OFF", vf->vf_id, vf_vsi->vsi_num, ret);
1370 else
1371 vf->spoofchk = ena;
1372
1373out_put_vf:
1374 ice_put_vf(vf);
1375 return ret;
1376}
1377
1378/**
1379 * ice_get_vf_cfg
1380 * @netdev: network interface device structure
1381 * @vf_id: VF identifier
1382 * @ivi: VF configuration structure
1383 *
1384 * return VF configuration
1385 */
1386int
1387ice_get_vf_cfg(struct net_device *netdev, int vf_id, struct ifla_vf_info *ivi)
1388{
1389 struct ice_pf *pf = ice_netdev_to_pf(netdev);
1390 struct ice_vf *vf;
1391 int ret;
1392
1393 vf = ice_get_vf_by_id(pf, vf_id);
1394 if (!vf)
1395 return -EINVAL;
1396
1397 ret = ice_check_vf_ready_for_cfg(vf);
1398 if (ret)
1399 goto out_put_vf;
1400
1401 ivi->vf = vf_id;
1402 ether_addr_copy(ivi->mac, vf->hw_lan_addr);
1403
1404 /* VF configuration for VLAN and applicable QoS */
1405 ivi->vlan = ice_vf_get_port_vlan_id(vf);
1406 ivi->qos = ice_vf_get_port_vlan_prio(vf);
1407 if (ice_vf_is_port_vlan_ena(vf))
1408 ivi->vlan_proto = cpu_to_be16(ice_vf_get_port_vlan_tpid(vf));
1409
1410 ivi->trusted = vf->trusted;
1411 ivi->spoofchk = vf->spoofchk;
1412 if (!vf->link_forced)
1413 ivi->linkstate = IFLA_VF_LINK_STATE_AUTO;
1414 else if (vf->link_up)
1415 ivi->linkstate = IFLA_VF_LINK_STATE_ENABLE;
1416 else
1417 ivi->linkstate = IFLA_VF_LINK_STATE_DISABLE;
1418 ivi->max_tx_rate = vf->max_tx_rate;
1419 ivi->min_tx_rate = vf->min_tx_rate;
1420
1421out_put_vf:
1422 ice_put_vf(vf);
1423 return ret;
1424}
1425
1426/**
1427 * ice_set_vf_mac
1428 * @netdev: network interface device structure
1429 * @vf_id: VF identifier
1430 * @mac: MAC address
1431 *
1432 * program VF MAC address
1433 */
1434int ice_set_vf_mac(struct net_device *netdev, int vf_id, u8 *mac)
1435{
1436 struct ice_pf *pf = ice_netdev_to_pf(netdev);
1437 struct ice_vf *vf;
1438 int ret;
1439
1440 if (is_multicast_ether_addr(mac)) {
1441 netdev_err(netdev, "%pM not a valid unicast address\n", mac);
1442 return -EINVAL;
1443 }
1444
1445 vf = ice_get_vf_by_id(pf, vf_id);
1446 if (!vf)
1447 return -EINVAL;
1448
1449 /* nothing left to do, unicast MAC already set */
1450 if (ether_addr_equal(vf->dev_lan_addr, mac) &&
1451 ether_addr_equal(vf->hw_lan_addr, mac)) {
1452 ret = 0;
1453 goto out_put_vf;
1454 }
1455
1456 ret = ice_check_vf_ready_for_cfg(vf);
1457 if (ret)
1458 goto out_put_vf;
1459
1460 mutex_lock(&vf->cfg_lock);
1461
1462 /* VF is notified of its new MAC via the PF's response to the
1463 * VIRTCHNL_OP_GET_VF_RESOURCES message after the VF has been reset
1464 */
1465 ether_addr_copy(vf->dev_lan_addr, mac);
1466 ether_addr_copy(vf->hw_lan_addr, mac);
1467 if (is_zero_ether_addr(mac)) {
1468 /* VF will send VIRTCHNL_OP_ADD_ETH_ADDR message with its MAC */
1469 vf->pf_set_mac = false;
1470 netdev_info(netdev, "Removing MAC on VF %d. VF driver will be reinitialized\n",
1471 vf->vf_id);
1472 } else {
1473 /* PF will add MAC rule for the VF */
1474 vf->pf_set_mac = true;
1475 netdev_info(netdev, "Setting MAC %pM on VF %d. VF driver will be reinitialized\n",
1476 mac, vf_id);
1477 }
1478
1479 ice_reset_vf(vf, ICE_VF_RESET_NOTIFY);
1480 mutex_unlock(&vf->cfg_lock);
1481
1482out_put_vf:
1483 ice_put_vf(vf);
1484 return ret;
1485}
1486
1487/**
1488 * ice_set_vf_trust
1489 * @netdev: network interface device structure
1490 * @vf_id: VF identifier
1491 * @trusted: Boolean value to enable/disable trusted VF
1492 *
1493 * Enable or disable a given VF as trusted
1494 */
1495int ice_set_vf_trust(struct net_device *netdev, int vf_id, bool trusted)
1496{
1497 struct ice_pf *pf = ice_netdev_to_pf(netdev);
1498 struct ice_vf *vf;
1499 int ret;
1500
1501 vf = ice_get_vf_by_id(pf, vf_id);
1502 if (!vf)
1503 return -EINVAL;
1504
1505 if (ice_is_eswitch_mode_switchdev(pf)) {
1506 dev_info(ice_pf_to_dev(pf), "Trusted VF is forbidden in switchdev mode\n");
1507 return -EOPNOTSUPP;
1508 }
1509
1510 ret = ice_check_vf_ready_for_cfg(vf);
1511 if (ret)
1512 goto out_put_vf;
1513
1514 /* Check if already trusted */
1515 if (trusted == vf->trusted) {
1516 ret = 0;
1517 goto out_put_vf;
1518 }
1519
1520 mutex_lock(&vf->cfg_lock);
1521
1522 vf->trusted = trusted;
1523 ice_reset_vf(vf, ICE_VF_RESET_NOTIFY);
1524 dev_info(ice_pf_to_dev(pf), "VF %u is now %strusted\n",
1525 vf_id, trusted ? "" : "un");
1526
1527 mutex_unlock(&vf->cfg_lock);
1528
1529out_put_vf:
1530 ice_put_vf(vf);
1531 return ret;
1532}
1533
1534/**
1535 * ice_set_vf_link_state
1536 * @netdev: network interface device structure
1537 * @vf_id: VF identifier
1538 * @link_state: required link state
1539 *
1540 * Set VF's link state, irrespective of physical link state status
1541 */
1542int ice_set_vf_link_state(struct net_device *netdev, int vf_id, int link_state)
1543{
1544 struct ice_pf *pf = ice_netdev_to_pf(netdev);
1545 struct ice_vf *vf;
1546 int ret;
1547
1548 vf = ice_get_vf_by_id(pf, vf_id);
1549 if (!vf)
1550 return -EINVAL;
1551
1552 ret = ice_check_vf_ready_for_cfg(vf);
1553 if (ret)
1554 goto out_put_vf;
1555
1556 switch (link_state) {
1557 case IFLA_VF_LINK_STATE_AUTO:
1558 vf->link_forced = false;
1559 break;
1560 case IFLA_VF_LINK_STATE_ENABLE:
1561 vf->link_forced = true;
1562 vf->link_up = true;
1563 break;
1564 case IFLA_VF_LINK_STATE_DISABLE:
1565 vf->link_forced = true;
1566 vf->link_up = false;
1567 break;
1568 default:
1569 ret = -EINVAL;
1570 goto out_put_vf;
1571 }
1572
1573 ice_vc_notify_vf_link_state(vf);
1574
1575out_put_vf:
1576 ice_put_vf(vf);
1577 return ret;
1578}
1579
1580/**
1581 * ice_calc_all_vfs_min_tx_rate - calculate cumulative min Tx rate on all VFs
1582 * @pf: PF associated with VFs
1583 */
1584static int ice_calc_all_vfs_min_tx_rate(struct ice_pf *pf)
1585{
1586 struct ice_vf *vf;
1587 unsigned int bkt;
1588 int rate = 0;
1589
1590 rcu_read_lock();
1591 ice_for_each_vf_rcu(pf, bkt, vf)
1592 rate += vf->min_tx_rate;
1593 rcu_read_unlock();
1594
1595 return rate;
1596}
1597
1598/**
1599 * ice_min_tx_rate_oversubscribed - check if min Tx rate causes oversubscription
1600 * @vf: VF trying to configure min_tx_rate
1601 * @min_tx_rate: min Tx rate in Mbps
1602 *
1603 * Check if the min_tx_rate being passed in will cause oversubscription of total
1604 * min_tx_rate based on the current link speed and all other VFs configured
1605 * min_tx_rate
1606 *
1607 * Return true if the passed min_tx_rate would cause oversubscription, else
1608 * return false
1609 */
1610static bool
1611ice_min_tx_rate_oversubscribed(struct ice_vf *vf, int min_tx_rate)
1612{
1613 struct ice_vsi *vsi = ice_get_vf_vsi(vf);
1614 int all_vfs_min_tx_rate;
1615 int link_speed_mbps;
1616
1617 if (WARN_ON(!vsi))
1618 return false;
1619
1620 link_speed_mbps = ice_get_link_speed_mbps(vsi);
1621 all_vfs_min_tx_rate = ice_calc_all_vfs_min_tx_rate(vf->pf);
1622
1623 /* this VF's previous rate is being overwritten */
1624 all_vfs_min_tx_rate -= vf->min_tx_rate;
1625
1626 if (all_vfs_min_tx_rate + min_tx_rate > link_speed_mbps) {
1627 dev_err(ice_pf_to_dev(vf->pf), "min_tx_rate of %d Mbps on VF %u would cause oversubscription of %d Mbps based on the current link speed %d Mbps\n",
1628 min_tx_rate, vf->vf_id,
1629 all_vfs_min_tx_rate + min_tx_rate - link_speed_mbps,
1630 link_speed_mbps);
1631 return true;
1632 }
1633
1634 return false;
1635}
1636
1637/**
1638 * ice_set_vf_bw - set min/max VF bandwidth
1639 * @netdev: network interface device structure
1640 * @vf_id: VF identifier
1641 * @min_tx_rate: Minimum Tx rate in Mbps
1642 * @max_tx_rate: Maximum Tx rate in Mbps
1643 */
1644int
1645ice_set_vf_bw(struct net_device *netdev, int vf_id, int min_tx_rate,
1646 int max_tx_rate)
1647{
1648 struct ice_pf *pf = ice_netdev_to_pf(netdev);
1649 struct ice_vsi *vsi;
1650 struct device *dev;
1651 struct ice_vf *vf;
1652 int ret;
1653
1654 dev = ice_pf_to_dev(pf);
1655
1656 vf = ice_get_vf_by_id(pf, vf_id);
1657 if (!vf)
1658 return -EINVAL;
1659
1660 ret = ice_check_vf_ready_for_cfg(vf);
1661 if (ret)
1662 goto out_put_vf;
1663
1664 vsi = ice_get_vf_vsi(vf);
1665 if (!vsi) {
1666 ret = -EINVAL;
1667 goto out_put_vf;
1668 }
1669
1670 if (min_tx_rate && ice_is_dcb_active(pf)) {
1671 dev_err(dev, "DCB on PF is currently enabled. VF min Tx rate limiting not allowed on this PF.\n");
1672 ret = -EOPNOTSUPP;
1673 goto out_put_vf;
1674 }
1675
1676 if (ice_min_tx_rate_oversubscribed(vf, min_tx_rate)) {
1677 ret = -EINVAL;
1678 goto out_put_vf;
1679 }
1680
1681 if (vf->min_tx_rate != (unsigned int)min_tx_rate) {
1682 ret = ice_set_min_bw_limit(vsi, (u64)min_tx_rate * 1000);
1683 if (ret) {
1684 dev_err(dev, "Unable to set min-tx-rate for VF %d\n",
1685 vf->vf_id);
1686 goto out_put_vf;
1687 }
1688
1689 vf->min_tx_rate = min_tx_rate;
1690 }
1691
1692 if (vf->max_tx_rate != (unsigned int)max_tx_rate) {
1693 ret = ice_set_max_bw_limit(vsi, (u64)max_tx_rate * 1000);
1694 if (ret) {
1695 dev_err(dev, "Unable to set max-tx-rate for VF %d\n",
1696 vf->vf_id);
1697 goto out_put_vf;
1698 }
1699
1700 vf->max_tx_rate = max_tx_rate;
1701 }
1702
1703out_put_vf:
1704 ice_put_vf(vf);
1705 return ret;
1706}
1707
1708/**
1709 * ice_get_vf_stats - populate some stats for the VF
1710 * @netdev: the netdev of the PF
1711 * @vf_id: the host OS identifier (0-255)
1712 * @vf_stats: pointer to the OS memory to be initialized
1713 */
1714int ice_get_vf_stats(struct net_device *netdev, int vf_id,
1715 struct ifla_vf_stats *vf_stats)
1716{
1717 struct ice_pf *pf = ice_netdev_to_pf(netdev);
1718 struct ice_eth_stats *stats;
1719 struct ice_vsi *vsi;
1720 struct ice_vf *vf;
1721 int ret;
1722
1723 vf = ice_get_vf_by_id(pf, vf_id);
1724 if (!vf)
1725 return -EINVAL;
1726
1727 ret = ice_check_vf_ready_for_cfg(vf);
1728 if (ret)
1729 goto out_put_vf;
1730
1731 vsi = ice_get_vf_vsi(vf);
1732 if (!vsi) {
1733 ret = -EINVAL;
1734 goto out_put_vf;
1735 }
1736
1737 ice_update_eth_stats(vsi);
1738 stats = &vsi->eth_stats;
1739
1740 memset(vf_stats, 0, sizeof(*vf_stats));
1741
1742 vf_stats->rx_packets = stats->rx_unicast + stats->rx_broadcast +
1743 stats->rx_multicast;
1744 vf_stats->tx_packets = stats->tx_unicast + stats->tx_broadcast +
1745 stats->tx_multicast;
1746 vf_stats->rx_bytes = stats->rx_bytes;
1747 vf_stats->tx_bytes = stats->tx_bytes;
1748 vf_stats->broadcast = stats->rx_broadcast;
1749 vf_stats->multicast = stats->rx_multicast;
1750 vf_stats->rx_dropped = stats->rx_discards;
1751 vf_stats->tx_dropped = stats->tx_discards;
1752
1753out_put_vf:
1754 ice_put_vf(vf);
1755 return ret;
1756}
1757
1758/**
1759 * ice_is_supported_port_vlan_proto - make sure the vlan_proto is supported
1760 * @hw: hardware structure used to check the VLAN mode
1761 * @vlan_proto: VLAN TPID being checked
1762 *
1763 * If the device is configured in Double VLAN Mode (DVM), then both ETH_P_8021Q
1764 * and ETH_P_8021AD are supported. If the device is configured in Single VLAN
1765 * Mode (SVM), then only ETH_P_8021Q is supported.
1766 */
1767static bool
1768ice_is_supported_port_vlan_proto(struct ice_hw *hw, u16 vlan_proto)
1769{
1770 bool is_supported = false;
1771
1772 switch (vlan_proto) {
1773 case ETH_P_8021Q:
1774 is_supported = true;
1775 break;
1776 case ETH_P_8021AD:
1777 if (ice_is_dvm_ena(hw))
1778 is_supported = true;
1779 break;
1780 }
1781
1782 return is_supported;
1783}
1784
1785/**
1786 * ice_set_vf_port_vlan
1787 * @netdev: network interface device structure
1788 * @vf_id: VF identifier
1789 * @vlan_id: VLAN ID being set
1790 * @qos: priority setting
1791 * @vlan_proto: VLAN protocol
1792 *
1793 * program VF Port VLAN ID and/or QoS
1794 */
1795int
1796ice_set_vf_port_vlan(struct net_device *netdev, int vf_id, u16 vlan_id, u8 qos,
1797 __be16 vlan_proto)
1798{
1799 struct ice_pf *pf = ice_netdev_to_pf(netdev);
1800 u16 local_vlan_proto = ntohs(vlan_proto);
1801 struct device *dev;
1802 struct ice_vf *vf;
1803 int ret;
1804
1805 dev = ice_pf_to_dev(pf);
1806
1807 if (vlan_id >= VLAN_N_VID || qos > 7) {
1808 dev_err(dev, "Invalid Port VLAN parameters for VF %d, ID %d, QoS %d\n",
1809 vf_id, vlan_id, qos);
1810 return -EINVAL;
1811 }
1812
1813 if (!ice_is_supported_port_vlan_proto(&pf->hw, local_vlan_proto)) {
1814 dev_err(dev, "VF VLAN protocol 0x%04x is not supported\n",
1815 local_vlan_proto);
1816 return -EPROTONOSUPPORT;
1817 }
1818
1819 vf = ice_get_vf_by_id(pf, vf_id);
1820 if (!vf)
1821 return -EINVAL;
1822
1823 ret = ice_check_vf_ready_for_cfg(vf);
1824 if (ret)
1825 goto out_put_vf;
1826
1827 if (ice_vf_get_port_vlan_prio(vf) == qos &&
1828 ice_vf_get_port_vlan_tpid(vf) == local_vlan_proto &&
1829 ice_vf_get_port_vlan_id(vf) == vlan_id) {
1830 /* duplicate request, so just return success */
1831 dev_dbg(dev, "Duplicate port VLAN %u, QoS %u, TPID 0x%04x request\n",
1832 vlan_id, qos, local_vlan_proto);
1833 ret = 0;
1834 goto out_put_vf;
1835 }
1836
1837 mutex_lock(&vf->cfg_lock);
1838
1839 vf->port_vlan_info = ICE_VLAN(local_vlan_proto, vlan_id, qos);
1840 if (ice_vf_is_port_vlan_ena(vf))
1841 dev_info(dev, "Setting VLAN %u, QoS %u, TPID 0x%04x on VF %d\n",
1842 vlan_id, qos, local_vlan_proto, vf_id);
1843 else
1844 dev_info(dev, "Clearing port VLAN on VF %d\n", vf_id);
1845
1846 ice_reset_vf(vf, ICE_VF_RESET_NOTIFY);
1847 mutex_unlock(&vf->cfg_lock);
1848
1849out_put_vf:
1850 ice_put_vf(vf);
1851 return ret;
1852}
1853
1854/**
1855 * ice_print_vf_rx_mdd_event - print VF Rx malicious driver detect event
1856 * @vf: pointer to the VF structure
1857 */
1858void ice_print_vf_rx_mdd_event(struct ice_vf *vf)
1859{
1860 struct ice_pf *pf = vf->pf;
1861 struct device *dev;
1862
1863 dev = ice_pf_to_dev(pf);
1864
1865 dev_info(dev, "%d Rx Malicious Driver Detection events detected on PF %d VF %d MAC %pM. mdd-auto-reset-vfs=%s\n",
1866 vf->mdd_rx_events.count, pf->hw.pf_id, vf->vf_id,
1867 vf->dev_lan_addr,
1868 test_bit(ICE_FLAG_MDD_AUTO_RESET_VF, pf->flags)
1869 ? "on" : "off");
1870}
1871
1872/**
1873 * ice_print_vfs_mdd_events - print VFs malicious driver detect event
1874 * @pf: pointer to the PF structure
1875 *
1876 * Called from ice_handle_mdd_event to rate limit and print VFs MDD events.
1877 */
1878void ice_print_vfs_mdd_events(struct ice_pf *pf)
1879{
1880 struct device *dev = ice_pf_to_dev(pf);
1881 struct ice_hw *hw = &pf->hw;
1882 struct ice_vf *vf;
1883 unsigned int bkt;
1884
1885 /* check that there are pending MDD events to print */
1886 if (!test_and_clear_bit(ICE_MDD_VF_PRINT_PENDING, pf->state))
1887 return;
1888
1889 /* VF MDD event logs are rate limited to one second intervals */
1890 if (time_is_after_jiffies(pf->vfs.last_printed_mdd_jiffies + HZ * 1))
1891 return;
1892
1893 pf->vfs.last_printed_mdd_jiffies = jiffies;
1894
1895 mutex_lock(&pf->vfs.table_lock);
1896 ice_for_each_vf(pf, bkt, vf) {
1897 /* only print Rx MDD event message if there are new events */
1898 if (vf->mdd_rx_events.count != vf->mdd_rx_events.last_printed) {
1899 vf->mdd_rx_events.last_printed =
1900 vf->mdd_rx_events.count;
1901 ice_print_vf_rx_mdd_event(vf);
1902 }
1903
1904 /* only print Tx MDD event message if there are new events */
1905 if (vf->mdd_tx_events.count != vf->mdd_tx_events.last_printed) {
1906 vf->mdd_tx_events.last_printed =
1907 vf->mdd_tx_events.count;
1908
1909 dev_info(dev, "%d Tx Malicious Driver Detection events detected on PF %d VF %d MAC %pM.\n",
1910 vf->mdd_tx_events.count, hw->pf_id, vf->vf_id,
1911 vf->dev_lan_addr);
1912 }
1913 }
1914 mutex_unlock(&pf->vfs.table_lock);
1915}
1916
1917/**
1918 * ice_restore_all_vfs_msi_state - restore VF MSI state after PF FLR
1919 * @pf: pointer to the PF structure
1920 *
1921 * Called when recovering from a PF FLR to restore interrupt capability to
1922 * the VFs.
1923 */
1924void ice_restore_all_vfs_msi_state(struct ice_pf *pf)
1925{
1926 struct ice_vf *vf;
1927 u32 bkt;
1928
1929 ice_for_each_vf(pf, bkt, vf)
1930 pci_restore_msi_state(vf->vfdev);
1931}
1// SPDX-License-Identifier: GPL-2.0
2/* Copyright (c) 2018, Intel Corporation. */
3
4#include "ice_common.h"
5#include "ice_sriov.h"
6
7/**
8 * ice_aq_send_msg_to_vf
9 * @hw: pointer to the hardware structure
10 * @vfid: VF ID to send msg
11 * @v_opcode: opcodes for VF-PF communication
12 * @v_retval: return error code
13 * @msg: pointer to the msg buffer
14 * @msglen: msg length
15 * @cd: pointer to command details
16 *
17 * Send message to VF driver (0x0802) using mailbox
18 * queue and asynchronously sending message via
19 * ice_sq_send_cmd() function
20 */
21enum ice_status
22ice_aq_send_msg_to_vf(struct ice_hw *hw, u16 vfid, u32 v_opcode, u32 v_retval,
23 u8 *msg, u16 msglen, struct ice_sq_cd *cd)
24{
25 struct ice_aqc_pf_vf_msg *cmd;
26 struct ice_aq_desc desc;
27
28 ice_fill_dflt_direct_cmd_desc(&desc, ice_mbx_opc_send_msg_to_vf);
29
30 cmd = &desc.params.virt;
31 cmd->id = cpu_to_le32(vfid);
32
33 desc.cookie_high = cpu_to_le32(v_opcode);
34 desc.cookie_low = cpu_to_le32(v_retval);
35
36 if (msglen)
37 desc.flags |= cpu_to_le16(ICE_AQ_FLAG_RD);
38
39 return ice_sq_send_cmd(hw, &hw->mailboxq, &desc, msg, msglen, cd);
40}
41
42/**
43 * ice_conv_link_speed_to_virtchnl
44 * @adv_link_support: determines the format of the returned link speed
45 * @link_speed: variable containing the link_speed to be converted
46 *
47 * Convert link speed supported by HW to link speed supported by virtchnl.
48 * If adv_link_support is true, then return link speed in Mbps. Else return
49 * link speed as a VIRTCHNL_LINK_SPEED_* casted to a u32. Note that the caller
50 * needs to cast back to an enum virtchnl_link_speed in the case where
51 * adv_link_support is false, but when adv_link_support is true the caller can
52 * expect the speed in Mbps.
53 */
54u32 ice_conv_link_speed_to_virtchnl(bool adv_link_support, u16 link_speed)
55{
56 u32 speed;
57
58 if (adv_link_support)
59 switch (link_speed) {
60 case ICE_AQ_LINK_SPEED_10MB:
61 speed = ICE_LINK_SPEED_10MBPS;
62 break;
63 case ICE_AQ_LINK_SPEED_100MB:
64 speed = ICE_LINK_SPEED_100MBPS;
65 break;
66 case ICE_AQ_LINK_SPEED_1000MB:
67 speed = ICE_LINK_SPEED_1000MBPS;
68 break;
69 case ICE_AQ_LINK_SPEED_2500MB:
70 speed = ICE_LINK_SPEED_2500MBPS;
71 break;
72 case ICE_AQ_LINK_SPEED_5GB:
73 speed = ICE_LINK_SPEED_5000MBPS;
74 break;
75 case ICE_AQ_LINK_SPEED_10GB:
76 speed = ICE_LINK_SPEED_10000MBPS;
77 break;
78 case ICE_AQ_LINK_SPEED_20GB:
79 speed = ICE_LINK_SPEED_20000MBPS;
80 break;
81 case ICE_AQ_LINK_SPEED_25GB:
82 speed = ICE_LINK_SPEED_25000MBPS;
83 break;
84 case ICE_AQ_LINK_SPEED_40GB:
85 speed = ICE_LINK_SPEED_40000MBPS;
86 break;
87 case ICE_AQ_LINK_SPEED_50GB:
88 speed = ICE_LINK_SPEED_50000MBPS;
89 break;
90 case ICE_AQ_LINK_SPEED_100GB:
91 speed = ICE_LINK_SPEED_100000MBPS;
92 break;
93 default:
94 speed = ICE_LINK_SPEED_UNKNOWN;
95 break;
96 }
97 else
98 /* Virtchnl speeds are not defined for every speed supported in
99 * the hardware. To maintain compatibility with older AVF
100 * drivers, while reporting the speed the new speed values are
101 * resolved to the closest known virtchnl speeds
102 */
103 switch (link_speed) {
104 case ICE_AQ_LINK_SPEED_10MB:
105 case ICE_AQ_LINK_SPEED_100MB:
106 speed = (u32)VIRTCHNL_LINK_SPEED_100MB;
107 break;
108 case ICE_AQ_LINK_SPEED_1000MB:
109 case ICE_AQ_LINK_SPEED_2500MB:
110 case ICE_AQ_LINK_SPEED_5GB:
111 speed = (u32)VIRTCHNL_LINK_SPEED_1GB;
112 break;
113 case ICE_AQ_LINK_SPEED_10GB:
114 speed = (u32)VIRTCHNL_LINK_SPEED_10GB;
115 break;
116 case ICE_AQ_LINK_SPEED_20GB:
117 speed = (u32)VIRTCHNL_LINK_SPEED_20GB;
118 break;
119 case ICE_AQ_LINK_SPEED_25GB:
120 speed = (u32)VIRTCHNL_LINK_SPEED_25GB;
121 break;
122 case ICE_AQ_LINK_SPEED_40GB:
123 case ICE_AQ_LINK_SPEED_50GB:
124 case ICE_AQ_LINK_SPEED_100GB:
125 speed = (u32)VIRTCHNL_LINK_SPEED_40GB;
126 break;
127 default:
128 speed = (u32)VIRTCHNL_LINK_SPEED_UNKNOWN;
129 break;
130 }
131
132 return speed;
133}
134
135/* The mailbox overflow detection algorithm helps to check if there
136 * is a possibility of a malicious VF transmitting too many MBX messages to the
137 * PF.
138 * 1. The mailbox snapshot structure, ice_mbx_snapshot, is initialized during
139 * driver initialization in ice_init_hw() using ice_mbx_init_snapshot().
140 * The struct ice_mbx_snapshot helps to track and traverse a static window of
141 * messages within the mailbox queue while looking for a malicious VF.
142 *
143 * 2. When the caller starts processing its mailbox queue in response to an
144 * interrupt, the structure ice_mbx_snapshot is expected to be cleared before
145 * the algorithm can be run for the first time for that interrupt. This can be
146 * done via ice_mbx_reset_snapshot().
147 *
148 * 3. For every message read by the caller from the MBX Queue, the caller must
149 * call the detection algorithm's entry function ice_mbx_vf_state_handler().
150 * Before every call to ice_mbx_vf_state_handler() the struct ice_mbx_data is
151 * filled as it is required to be passed to the algorithm.
152 *
153 * 4. Every time a message is read from the MBX queue, a VFId is received which
154 * is passed to the state handler. The boolean output is_malvf of the state
155 * handler ice_mbx_vf_state_handler() serves as an indicator to the caller
156 * whether this VF is malicious or not.
157 *
158 * 5. When a VF is identified to be malicious, the caller can send a message
159 * to the system administrator. The caller can invoke ice_mbx_report_malvf()
160 * to help determine if a malicious VF is to be reported or not. This function
161 * requires the caller to maintain a global bitmap to track all malicious VFs
162 * and pass that to ice_mbx_report_malvf() along with the VFID which was identified
163 * to be malicious by ice_mbx_vf_state_handler().
164 *
165 * 6. The global bitmap maintained by PF can be cleared completely if PF is in
166 * reset or the bit corresponding to a VF can be cleared if that VF is in reset.
167 * When a VF is shut down and brought back up, we assume that the new VF
168 * brought up is not malicious and hence report it if found malicious.
169 *
170 * 7. The function ice_mbx_reset_snapshot() is called to reset the information
171 * in ice_mbx_snapshot for every new mailbox interrupt handled.
172 *
173 * 8. The memory allocated for variables in ice_mbx_snapshot is de-allocated
174 * when driver is unloaded.
175 */
176#define ICE_RQ_DATA_MASK(rq_data) ((rq_data) & PF_MBX_ARQH_ARQH_M)
177/* Using the highest value for an unsigned 16-bit value 0xFFFF to indicate that
178 * the max messages check must be ignored in the algorithm
179 */
180#define ICE_IGNORE_MAX_MSG_CNT 0xFFFF
181
182/**
183 * ice_mbx_traverse - Pass through mailbox snapshot
184 * @hw: pointer to the HW struct
185 * @new_state: new algorithm state
186 *
187 * Traversing the mailbox static snapshot without checking
188 * for malicious VFs.
189 */
190static void
191ice_mbx_traverse(struct ice_hw *hw,
192 enum ice_mbx_snapshot_state *new_state)
193{
194 struct ice_mbx_snap_buffer_data *snap_buf;
195 u32 num_iterations;
196
197 snap_buf = &hw->mbx_snapshot.mbx_buf;
198
199 /* As mailbox buffer is circular, applying a mask
200 * on the incremented iteration count.
201 */
202 num_iterations = ICE_RQ_DATA_MASK(++snap_buf->num_iterations);
203
204 /* Checking either of the below conditions to exit snapshot traversal:
205 * Condition-1: If the number of iterations in the mailbox is equal to
206 * the mailbox head which would indicate that we have reached the end
207 * of the static snapshot.
208 * Condition-2: If the maximum messages serviced in the mailbox for a
209 * given interrupt is the highest possible value then there is no need
210 * to check if the number of messages processed is equal to it. If not
211 * check if the number of messages processed is greater than or equal
212 * to the maximum number of mailbox entries serviced in current work item.
213 */
214 if (num_iterations == snap_buf->head ||
215 (snap_buf->max_num_msgs_mbx < ICE_IGNORE_MAX_MSG_CNT &&
216 ++snap_buf->num_msg_proc >= snap_buf->max_num_msgs_mbx))
217 *new_state = ICE_MAL_VF_DETECT_STATE_NEW_SNAPSHOT;
218}
219
220/**
221 * ice_mbx_detect_malvf - Detect malicious VF in snapshot
222 * @hw: pointer to the HW struct
223 * @vf_id: relative virtual function ID
224 * @new_state: new algorithm state
225 * @is_malvf: boolean output to indicate if VF is malicious
226 *
227 * This function tracks the number of asynchronous messages
228 * sent per VF and marks the VF as malicious if it exceeds
229 * the permissible number of messages to send.
230 */
231static enum ice_status
232ice_mbx_detect_malvf(struct ice_hw *hw, u16 vf_id,
233 enum ice_mbx_snapshot_state *new_state,
234 bool *is_malvf)
235{
236 struct ice_mbx_snapshot *snap = &hw->mbx_snapshot;
237
238 if (vf_id >= snap->mbx_vf.vfcntr_len)
239 return ICE_ERR_OUT_OF_RANGE;
240
241 /* increment the message count in the VF array */
242 snap->mbx_vf.vf_cntr[vf_id]++;
243
244 if (snap->mbx_vf.vf_cntr[vf_id] >= ICE_ASYNC_VF_MSG_THRESHOLD)
245 *is_malvf = true;
246
247 /* continue to iterate through the mailbox snapshot */
248 ice_mbx_traverse(hw, new_state);
249
250 return 0;
251}
252
253/**
254 * ice_mbx_reset_snapshot - Reset mailbox snapshot structure
255 * @snap: pointer to mailbox snapshot structure in the ice_hw struct
256 *
257 * Reset the mailbox snapshot structure and clear VF counter array.
258 */
259static void ice_mbx_reset_snapshot(struct ice_mbx_snapshot *snap)
260{
261 u32 vfcntr_len;
262
263 if (!snap || !snap->mbx_vf.vf_cntr)
264 return;
265
266 /* Clear VF counters. */
267 vfcntr_len = snap->mbx_vf.vfcntr_len;
268 if (vfcntr_len)
269 memset(snap->mbx_vf.vf_cntr, 0,
270 (vfcntr_len * sizeof(*snap->mbx_vf.vf_cntr)));
271
272 /* Reset mailbox snapshot for a new capture. */
273 memset(&snap->mbx_buf, 0, sizeof(snap->mbx_buf));
274 snap->mbx_buf.state = ICE_MAL_VF_DETECT_STATE_NEW_SNAPSHOT;
275}
276
277/**
278 * ice_mbx_vf_state_handler - Handle states of the overflow algorithm
279 * @hw: pointer to the HW struct
280 * @mbx_data: pointer to structure containing mailbox data
281 * @vf_id: relative virtual function (VF) ID
282 * @is_malvf: boolean output to indicate if VF is malicious
283 *
284 * The function serves as an entry point for the malicious VF
285 * detection algorithm by handling the different states and state
286 * transitions of the algorithm:
287 * New snapshot: This state is entered when creating a new static
288 * snapshot. The data from any previous mailbox snapshot is
289 * cleared and a new capture of the mailbox head and tail is
290 * logged. This will be the new static snapshot to detect
291 * asynchronous messages sent by VFs. On capturing the snapshot
292 * and depending on whether the number of pending messages in that
293 * snapshot exceed the watermark value, the state machine enters
294 * traverse or detect states.
295 * Traverse: If pending message count is below watermark then iterate
296 * through the snapshot without any action on VF.
297 * Detect: If pending message count exceeds watermark traverse
298 * the static snapshot and look for a malicious VF.
299 */
300enum ice_status
301ice_mbx_vf_state_handler(struct ice_hw *hw,
302 struct ice_mbx_data *mbx_data, u16 vf_id,
303 bool *is_malvf)
304{
305 struct ice_mbx_snapshot *snap = &hw->mbx_snapshot;
306 struct ice_mbx_snap_buffer_data *snap_buf;
307 struct ice_ctl_q_info *cq = &hw->mailboxq;
308 enum ice_mbx_snapshot_state new_state;
309 enum ice_status status = 0;
310
311 if (!is_malvf || !mbx_data)
312 return ICE_ERR_BAD_PTR;
313
314 /* When entering the mailbox state machine assume that the VF
315 * is not malicious until detected.
316 */
317 *is_malvf = false;
318
319 /* Checking if max messages allowed to be processed while servicing current
320 * interrupt is not less than the defined AVF message threshold.
321 */
322 if (mbx_data->max_num_msgs_mbx <= ICE_ASYNC_VF_MSG_THRESHOLD)
323 return ICE_ERR_INVAL_SIZE;
324
325 /* The watermark value should not be lesser than the threshold limit
326 * set for the number of asynchronous messages a VF can send to mailbox
327 * nor should it be greater than the maximum number of messages in the
328 * mailbox serviced in current interrupt.
329 */
330 if (mbx_data->async_watermark_val < ICE_ASYNC_VF_MSG_THRESHOLD ||
331 mbx_data->async_watermark_val > mbx_data->max_num_msgs_mbx)
332 return ICE_ERR_PARAM;
333
334 new_state = ICE_MAL_VF_DETECT_STATE_INVALID;
335 snap_buf = &snap->mbx_buf;
336
337 switch (snap_buf->state) {
338 case ICE_MAL_VF_DETECT_STATE_NEW_SNAPSHOT:
339 /* Clear any previously held data in mailbox snapshot structure. */
340 ice_mbx_reset_snapshot(snap);
341
342 /* Collect the pending ARQ count, number of messages processed and
343 * the maximum number of messages allowed to be processed from the
344 * Mailbox for current interrupt.
345 */
346 snap_buf->num_pending_arq = mbx_data->num_pending_arq;
347 snap_buf->num_msg_proc = mbx_data->num_msg_proc;
348 snap_buf->max_num_msgs_mbx = mbx_data->max_num_msgs_mbx;
349
350 /* Capture a new static snapshot of the mailbox by logging the
351 * head and tail of snapshot and set num_iterations to the tail
352 * value to mark the start of the iteration through the snapshot.
353 */
354 snap_buf->head = ICE_RQ_DATA_MASK(cq->rq.next_to_clean +
355 mbx_data->num_pending_arq);
356 snap_buf->tail = ICE_RQ_DATA_MASK(cq->rq.next_to_clean - 1);
357 snap_buf->num_iterations = snap_buf->tail;
358
359 /* Pending ARQ messages returned by ice_clean_rq_elem
360 * is the difference between the head and tail of the
361 * mailbox queue. Comparing this value against the watermark
362 * helps to check if we potentially have malicious VFs.
363 */
364 if (snap_buf->num_pending_arq >=
365 mbx_data->async_watermark_val) {
366 new_state = ICE_MAL_VF_DETECT_STATE_DETECT;
367 status = ice_mbx_detect_malvf(hw, vf_id, &new_state, is_malvf);
368 } else {
369 new_state = ICE_MAL_VF_DETECT_STATE_TRAVERSE;
370 ice_mbx_traverse(hw, &new_state);
371 }
372 break;
373
374 case ICE_MAL_VF_DETECT_STATE_TRAVERSE:
375 new_state = ICE_MAL_VF_DETECT_STATE_TRAVERSE;
376 ice_mbx_traverse(hw, &new_state);
377 break;
378
379 case ICE_MAL_VF_DETECT_STATE_DETECT:
380 new_state = ICE_MAL_VF_DETECT_STATE_DETECT;
381 status = ice_mbx_detect_malvf(hw, vf_id, &new_state, is_malvf);
382 break;
383
384 default:
385 new_state = ICE_MAL_VF_DETECT_STATE_INVALID;
386 status = ICE_ERR_CFG;
387 }
388
389 snap_buf->state = new_state;
390
391 return status;
392}
393
394/**
395 * ice_mbx_report_malvf - Track and note malicious VF
396 * @hw: pointer to the HW struct
397 * @all_malvfs: all malicious VFs tracked by PF
398 * @bitmap_len: length of bitmap in bits
399 * @vf_id: relative virtual function ID of the malicious VF
400 * @report_malvf: boolean to indicate if malicious VF must be reported
401 *
402 * This function will update a bitmap that keeps track of the malicious
403 * VFs attached to the PF. A malicious VF must be reported only once if
404 * discovered between VF resets or loading so the function checks
405 * the input vf_id against the bitmap to verify if the VF has been
406 * detected in any previous mailbox iterations.
407 */
408enum ice_status
409ice_mbx_report_malvf(struct ice_hw *hw, unsigned long *all_malvfs,
410 u16 bitmap_len, u16 vf_id, bool *report_malvf)
411{
412 if (!all_malvfs || !report_malvf)
413 return ICE_ERR_PARAM;
414
415 *report_malvf = false;
416
417 if (bitmap_len < hw->mbx_snapshot.mbx_vf.vfcntr_len)
418 return ICE_ERR_INVAL_SIZE;
419
420 if (vf_id >= bitmap_len)
421 return ICE_ERR_OUT_OF_RANGE;
422
423 /* If the vf_id is found in the bitmap set bit and boolean to true */
424 if (!test_and_set_bit(vf_id, all_malvfs))
425 *report_malvf = true;
426
427 return 0;
428}
429
430/**
431 * ice_mbx_clear_malvf - Clear VF bitmap and counter for VF ID
432 * @snap: pointer to the mailbox snapshot structure
433 * @all_malvfs: all malicious VFs tracked by PF
434 * @bitmap_len: length of bitmap in bits
435 * @vf_id: relative virtual function ID of the malicious VF
436 *
437 * In case of a VF reset, this function can be called to clear
438 * the bit corresponding to the VF ID in the bitmap tracking all
439 * malicious VFs attached to the PF. The function also clears the
440 * VF counter array at the index of the VF ID. This is to ensure
441 * that the new VF loaded is not considered malicious before going
442 * through the overflow detection algorithm.
443 */
444enum ice_status
445ice_mbx_clear_malvf(struct ice_mbx_snapshot *snap, unsigned long *all_malvfs,
446 u16 bitmap_len, u16 vf_id)
447{
448 if (!snap || !all_malvfs)
449 return ICE_ERR_PARAM;
450
451 if (bitmap_len < snap->mbx_vf.vfcntr_len)
452 return ICE_ERR_INVAL_SIZE;
453
454 /* Ensure VF ID value is not larger than bitmap or VF counter length */
455 if (vf_id >= bitmap_len || vf_id >= snap->mbx_vf.vfcntr_len)
456 return ICE_ERR_OUT_OF_RANGE;
457
458 /* Clear VF ID bit in the bitmap tracking malicious VFs attached to PF */
459 clear_bit(vf_id, all_malvfs);
460
461 /* Clear the VF counter in the mailbox snapshot structure for that VF ID.
462 * This is to ensure that if a VF is unloaded and a new one brought back
463 * up with the same VF ID for a snapshot currently in traversal or detect
464 * state the counter for that VF ID does not increment on top of existing
465 * values in the mailbox overflow detection algorithm.
466 */
467 snap->mbx_vf.vf_cntr[vf_id] = 0;
468
469 return 0;
470}
471
472/**
473 * ice_mbx_init_snapshot - Initialize mailbox snapshot structure
474 * @hw: pointer to the hardware structure
475 * @vf_count: number of VFs allocated on a PF
476 *
477 * Clear the mailbox snapshot structure and allocate memory
478 * for the VF counter array based on the number of VFs allocated
479 * on that PF.
480 *
481 * Assumption: This function will assume ice_get_caps() has already been
482 * called to ensure that the vf_count can be compared against the number
483 * of VFs supported as defined in the functional capabilities of the device.
484 */
485enum ice_status ice_mbx_init_snapshot(struct ice_hw *hw, u16 vf_count)
486{
487 struct ice_mbx_snapshot *snap = &hw->mbx_snapshot;
488
489 /* Ensure that the number of VFs allocated is non-zero and
490 * is not greater than the number of supported VFs defined in
491 * the functional capabilities of the PF.
492 */
493 if (!vf_count || vf_count > hw->func_caps.num_allocd_vfs)
494 return ICE_ERR_INVAL_SIZE;
495
496 snap->mbx_vf.vf_cntr = devm_kcalloc(ice_hw_to_dev(hw), vf_count,
497 sizeof(*snap->mbx_vf.vf_cntr),
498 GFP_KERNEL);
499 if (!snap->mbx_vf.vf_cntr)
500 return ICE_ERR_NO_MEMORY;
501
502 /* Setting the VF counter length to the number of allocated
503 * VFs for given PF's functional capabilities.
504 */
505 snap->mbx_vf.vfcntr_len = vf_count;
506
507 /* Clear mbx_buf in the mailbox snaphot structure and setting the
508 * mailbox snapshot state to a new capture.
509 */
510 memset(&snap->mbx_buf, 0, sizeof(snap->mbx_buf));
511 snap->mbx_buf.state = ICE_MAL_VF_DETECT_STATE_NEW_SNAPSHOT;
512
513 return 0;
514}
515
516/**
517 * ice_mbx_deinit_snapshot - Free mailbox snapshot structure
518 * @hw: pointer to the hardware structure
519 *
520 * Clear the mailbox snapshot structure and free the VF counter array.
521 */
522void ice_mbx_deinit_snapshot(struct ice_hw *hw)
523{
524 struct ice_mbx_snapshot *snap = &hw->mbx_snapshot;
525
526 /* Free VF counter array and reset VF counter length */
527 devm_kfree(ice_hw_to_dev(hw), snap->mbx_vf.vf_cntr);
528 snap->mbx_vf.vfcntr_len = 0;
529
530 /* Clear mbx_buf in the mailbox snaphot structure */
531 memset(&snap->mbx_buf, 0, sizeof(snap->mbx_buf));
532}