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