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 "siena_sriov.h"
20#include "vfdi.h"
21
22/* Number of longs required to track all the VIs in a VF */
23#define VI_MASK_LENGTH BITS_TO_LONGS(1 << EFX_VI_SCALE_MAX)
24
25/* Maximum number of RX queues supported */
26#define VF_MAX_RX_QUEUES 63
27
28/**
29 * enum efx_vf_tx_filter_mode - TX MAC filtering behaviour
30 * @VF_TX_FILTER_OFF: Disabled
31 * @VF_TX_FILTER_AUTO: Enabled if MAC address assigned to VF and only
32 *	2 TX queues allowed per VF.
33 * @VF_TX_FILTER_ON: Enabled
34 */
35enum efx_vf_tx_filter_mode {
36	VF_TX_FILTER_OFF,
37	VF_TX_FILTER_AUTO,
38	VF_TX_FILTER_ON,
39};
40
41/**
42 * struct siena_vf - Back-end resource and protocol state for a PCI VF
43 * @efx: The Efx NIC owning this VF
44 * @pci_rid: The PCI requester ID for this VF
45 * @pci_name: The PCI name (formatted address) of this VF
46 * @index: Index of VF within its port and PF.
47 * @req: VFDI incoming request work item. Incoming USR_EV events are received
48 *	by the NAPI handler, but must be handled by executing MCDI requests
49 *	inside a work item.
50 * @req_addr: VFDI incoming request DMA address (in VF's PCI address space).
51 * @req_type: Expected next incoming (from VF) %VFDI_EV_TYPE member.
52 * @req_seqno: Expected next incoming (from VF) %VFDI_EV_SEQ member.
53 * @msg_seqno: Next %VFDI_EV_SEQ member to reply to VF. Protected by
54 *	@status_lock
55 * @busy: VFDI request queued to be processed or being processed. Receiving
56 *	a VFDI request when @busy is set is an error condition.
57 * @buf: Incoming VFDI requests are DMA from the VF into this buffer.
58 * @buftbl_base: Buffer table entries for this VF start at this index.
59 * @rx_filtering: Receive filtering has been requested by the VF driver.
60 * @rx_filter_flags: The flags sent in the %VFDI_OP_INSERT_FILTER request.
61 * @rx_filter_qid: VF relative qid for RX filter requested by VF.
62 * @rx_filter_id: Receive MAC filter ID. Only one filter per VF is supported.
63 * @tx_filter_mode: Transmit MAC filtering mode.
64 * @tx_filter_id: Transmit MAC filter ID.
65 * @addr: The MAC address and outer vlan tag of the VF.
66 * @status_addr: VF DMA address of page for &struct vfdi_status updates.
67 * @status_lock: Mutex protecting @msg_seqno, @status_addr, @addr,
68 *	@peer_page_addrs and @peer_page_count from simultaneous
69 *	updates by the VM and consumption by
70 *	efx_siena_sriov_update_vf_addr()
71 * @peer_page_addrs: Pointer to an array of guest pages for local addresses.
72 * @peer_page_count: Number of entries in @peer_page_count.
73 * @evq0_addrs: Array of guest pages backing evq0.
74 * @evq0_count: Number of entries in @evq0_addrs.
75 * @flush_waitq: wait queue used by %VFDI_OP_FINI_ALL_QUEUES handler
76 *	to wait for flush completions.
77 * @txq_lock: Mutex for TX queue allocation.
78 * @txq_mask: Mask of initialized transmit queues.
79 * @txq_count: Number of initialized transmit queues.
80 * @rxq_mask: Mask of initialized receive queues.
81 * @rxq_count: Number of initialized receive queues.
82 * @rxq_retry_mask: Mask or receive queues that need to be flushed again
83 *	due to flush failure.
84 * @rxq_retry_count: Number of receive queues in @rxq_retry_mask.
85 * @reset_work: Work item to schedule a VF reset.
86 */
87struct siena_vf {
88	struct efx_nic *efx;
89	unsigned int pci_rid;
90	char pci_name[13]; /* dddd:bb:dd.f */
91	unsigned int index;
92	struct work_struct req;
93	u64 req_addr;
94	int req_type;
95	unsigned req_seqno;
96	unsigned msg_seqno;
97	bool busy;
98	struct efx_buffer buf;
99	unsigned buftbl_base;
100	bool rx_filtering;
101	enum efx_filter_flags rx_filter_flags;
102	unsigned rx_filter_qid;
103	int rx_filter_id;
104	enum efx_vf_tx_filter_mode tx_filter_mode;
105	int tx_filter_id;
106	struct vfdi_endpoint addr;
107	u64 status_addr;
108	struct mutex status_lock;
109	u64 *peer_page_addrs;
110	unsigned peer_page_count;
111	u64 evq0_addrs[EFX_MAX_VF_EVQ_SIZE * sizeof(efx_qword_t) /
112		       EFX_BUF_SIZE];
113	unsigned evq0_count;
114	wait_queue_head_t flush_waitq;
115	struct mutex txq_lock;
116	unsigned long txq_mask[VI_MASK_LENGTH];
117	unsigned txq_count;
118	unsigned long rxq_mask[VI_MASK_LENGTH];
119	unsigned rxq_count;
120	unsigned long rxq_retry_mask[VI_MASK_LENGTH];
121	atomic_t rxq_retry_count;
122	struct work_struct reset_work;
123};
124
125struct efx_memcpy_req {
126	unsigned int from_rid;
127	void *from_buf;
128	u64 from_addr;
129	unsigned int to_rid;
130	u64 to_addr;
131	unsigned length;
132};
133
134/**
135 * struct efx_local_addr - A MAC address on the vswitch without a VF.
136 *
137 * Siena does not have a switch, so VFs can't transmit data to each
138 * other. Instead the VFs must be made aware of the local addresses
139 * on the vswitch, so that they can arrange for an alternative
140 * software datapath to be used.
141 *
142 * @link: List head for insertion into efx->local_addr_list.
143 * @addr: Ethernet address
144 */
145struct efx_local_addr {
146	struct list_head link;
147	u8 addr[ETH_ALEN];
148};
149
150/**
151 * struct efx_endpoint_page - Page of vfdi_endpoint structures
152 *
153 * @link: List head for insertion into efx->local_page_list.
154 * @ptr: Pointer to page.
155 * @addr: DMA address of page.
156 */
157struct efx_endpoint_page {
158	struct list_head link;
159	void *ptr;
160	dma_addr_t addr;
161};
162
163/* Buffer table entries are reserved txq0,rxq0,evq0,txq1,rxq1,evq1 */
164#define EFX_BUFTBL_TXQ_BASE(_vf, _qid)					\
165	((_vf)->buftbl_base + EFX_VF_BUFTBL_PER_VI * (_qid))
166#define EFX_BUFTBL_RXQ_BASE(_vf, _qid)					\
167	(EFX_BUFTBL_TXQ_BASE(_vf, _qid) +				\
168	 (EFX_MAX_DMAQ_SIZE * sizeof(efx_qword_t) / EFX_BUF_SIZE))
169#define EFX_BUFTBL_EVQ_BASE(_vf, _qid)					\
170	(EFX_BUFTBL_TXQ_BASE(_vf, _qid) +				\
171	 (2 * EFX_MAX_DMAQ_SIZE * sizeof(efx_qword_t) / EFX_BUF_SIZE))
172
173#define EFX_FIELD_MASK(_field)			\
174	((1 << _field ## _WIDTH) - 1)
175
176/* VFs can only use this many transmit channels */
177static unsigned int vf_max_tx_channels = 2;
178module_param(vf_max_tx_channels, uint, 0444);
179MODULE_PARM_DESC(vf_max_tx_channels,
180		 "Limit the number of TX channels VFs can use");
181
182static int max_vfs = -1;
183module_param(max_vfs, int, 0444);
184MODULE_PARM_DESC(max_vfs,
185		 "Reduce the number of VFs initialized by the driver");
186
187/* Workqueue used by VFDI communication.  We can't use the global
188 * workqueue because it may be running the VF driver's probe()
189 * routine, which will be blocked there waiting for a VFDI response.
190 */
191static struct workqueue_struct *vfdi_workqueue;
192
193static unsigned abs_index(struct siena_vf *vf, unsigned index)
194{
195	return EFX_VI_BASE + vf->index * efx_vf_size(vf->efx) + index;
196}
197
198static int efx_siena_sriov_cmd(struct efx_nic *efx, bool enable,
199			       unsigned *vi_scale_out, unsigned *vf_total_out)
200{
201	MCDI_DECLARE_BUF(inbuf, MC_CMD_SRIOV_IN_LEN);
202	MCDI_DECLARE_BUF(outbuf, MC_CMD_SRIOV_OUT_LEN);
203	unsigned vi_scale, vf_total;
204	size_t outlen;
205	int rc;
206
207	MCDI_SET_DWORD(inbuf, SRIOV_IN_ENABLE, enable ? 1 : 0);
208	MCDI_SET_DWORD(inbuf, SRIOV_IN_VI_BASE, EFX_VI_BASE);
209	MCDI_SET_DWORD(inbuf, SRIOV_IN_VF_COUNT, efx->vf_count);
210
211	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_SRIOV, inbuf, MC_CMD_SRIOV_IN_LEN,
212				outbuf, MC_CMD_SRIOV_OUT_LEN, &outlen);
213	if (rc)
214		return rc;
215	if (outlen < MC_CMD_SRIOV_OUT_LEN)
216		return -EIO;
217
218	vf_total = MCDI_DWORD(outbuf, SRIOV_OUT_VF_TOTAL);
219	vi_scale = MCDI_DWORD(outbuf, SRIOV_OUT_VI_SCALE);
220	if (vi_scale > EFX_VI_SCALE_MAX)
221		return -EOPNOTSUPP;
222
223	if (vi_scale_out)
224		*vi_scale_out = vi_scale;
225	if (vf_total_out)
226		*vf_total_out = vf_total;
227
228	return 0;
229}
230
231static void efx_siena_sriov_usrev(struct efx_nic *efx, bool enabled)
232{
233	struct siena_nic_data *nic_data = efx->nic_data;
234	efx_oword_t reg;
235
236	EFX_POPULATE_OWORD_2(reg,
237			     FRF_CZ_USREV_DIS, enabled ? 0 : 1,
238			     FRF_CZ_DFLT_EVQ, nic_data->vfdi_channel->channel);
239	efx_writeo(efx, &reg, FR_CZ_USR_EV_CFG);
240}
241
242static int efx_siena_sriov_memcpy(struct efx_nic *efx,
243				  struct efx_memcpy_req *req,
244				  unsigned int count)
245{
246	MCDI_DECLARE_BUF(inbuf, MCDI_CTL_SDU_LEN_MAX_V1);
247	MCDI_DECLARE_STRUCT_PTR(record);
248	unsigned int index, used;
249	u64 from_addr;
250	u32 from_rid;
251	int rc;
252
253	mb();	/* Finish writing source/reading dest before DMA starts */
254
255	if (WARN_ON(count > MC_CMD_MEMCPY_IN_RECORD_MAXNUM))
256		return -ENOBUFS;
257	used = MC_CMD_MEMCPY_IN_LEN(count);
258
259	for (index = 0; index < count; index++) {
260		record = MCDI_ARRAY_STRUCT_PTR(inbuf, MEMCPY_IN_RECORD, index);
261		MCDI_SET_DWORD(record, MEMCPY_RECORD_TYPEDEF_NUM_RECORDS,
262			       count);
263		MCDI_SET_DWORD(record, MEMCPY_RECORD_TYPEDEF_TO_RID,
264			       req->to_rid);
265		MCDI_SET_QWORD(record, MEMCPY_RECORD_TYPEDEF_TO_ADDR,
266			       req->to_addr);
267		if (req->from_buf == NULL) {
268			from_rid = req->from_rid;
269			from_addr = req->from_addr;
270		} else {
271			if (WARN_ON(used + req->length >
272				    MCDI_CTL_SDU_LEN_MAX_V1)) {
273				rc = -ENOBUFS;
274				goto out;
275			}
276
277			from_rid = MC_CMD_MEMCPY_RECORD_TYPEDEF_RID_INLINE;
278			from_addr = used;
279			memcpy(_MCDI_PTR(inbuf, used), req->from_buf,
280			       req->length);
281			used += req->length;
282		}
283
284		MCDI_SET_DWORD(record, MEMCPY_RECORD_TYPEDEF_FROM_RID, from_rid);
285		MCDI_SET_QWORD(record, MEMCPY_RECORD_TYPEDEF_FROM_ADDR,
286			       from_addr);
287		MCDI_SET_DWORD(record, MEMCPY_RECORD_TYPEDEF_LENGTH,
288			       req->length);
289
290		++req;
291	}
292
293	rc = efx_mcdi_rpc(efx, MC_CMD_MEMCPY, inbuf, used, NULL, 0, NULL);
294out:
295	mb();	/* Don't write source/read dest before DMA is complete */
296
297	return rc;
298}
299
300/* The TX filter is entirely controlled by this driver, and is modified
301 * underneath the feet of the VF
302 */
303static void efx_siena_sriov_reset_tx_filter(struct siena_vf *vf)
304{
305	struct efx_nic *efx = vf->efx;
306	struct efx_filter_spec filter;
307	u16 vlan;
308	int rc;
309
310	if (vf->tx_filter_id != -1) {
311		efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED,
312					  vf->tx_filter_id);
313		netif_dbg(efx, hw, efx->net_dev, "Removed vf %s tx filter %d\n",
314			  vf->pci_name, vf->tx_filter_id);
315		vf->tx_filter_id = -1;
316	}
317
318	if (is_zero_ether_addr(vf->addr.mac_addr))
319		return;
320
321	/* Turn on TX filtering automatically if not explicitly
322	 * enabled or disabled.
323	 */
324	if (vf->tx_filter_mode == VF_TX_FILTER_AUTO && vf_max_tx_channels <= 2)
325		vf->tx_filter_mode = VF_TX_FILTER_ON;
326
327	vlan = ntohs(vf->addr.tci) & VLAN_VID_MASK;
328	efx_filter_init_tx(&filter, abs_index(vf, 0));
329	rc = efx_filter_set_eth_local(&filter,
330				      vlan ? vlan : EFX_FILTER_VID_UNSPEC,
331				      vf->addr.mac_addr);
332	BUG_ON(rc);
333
334	rc = efx_filter_insert_filter(efx, &filter, true);
335	if (rc < 0) {
336		netif_warn(efx, hw, efx->net_dev,
337			   "Unable to migrate tx filter for vf %s\n",
338			   vf->pci_name);
339	} else {
340		netif_dbg(efx, hw, efx->net_dev, "Inserted vf %s tx filter %d\n",
341			  vf->pci_name, rc);
342		vf->tx_filter_id = rc;
343	}
344}
345
346/* The RX filter is managed here on behalf of the VF driver */
347static void efx_siena_sriov_reset_rx_filter(struct siena_vf *vf)
348{
349	struct efx_nic *efx = vf->efx;
350	struct efx_filter_spec filter;
351	u16 vlan;
352	int rc;
353
354	if (vf->rx_filter_id != -1) {
355		efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED,
356					  vf->rx_filter_id);
357		netif_dbg(efx, hw, efx->net_dev, "Removed vf %s rx filter %d\n",
358			  vf->pci_name, vf->rx_filter_id);
359		vf->rx_filter_id = -1;
360	}
361
362	if (!vf->rx_filtering || is_zero_ether_addr(vf->addr.mac_addr))
363		return;
364
365	vlan = ntohs(vf->addr.tci) & VLAN_VID_MASK;
366	efx_filter_init_rx(&filter, EFX_FILTER_PRI_REQUIRED,
367			   vf->rx_filter_flags,
368			   abs_index(vf, vf->rx_filter_qid));
369	rc = efx_filter_set_eth_local(&filter,
370				      vlan ? vlan : EFX_FILTER_VID_UNSPEC,
371				      vf->addr.mac_addr);
372	BUG_ON(rc);
373
374	rc = efx_filter_insert_filter(efx, &filter, true);
375	if (rc < 0) {
376		netif_warn(efx, hw, efx->net_dev,
377			   "Unable to insert rx filter for vf %s\n",
378			   vf->pci_name);
379	} else {
380		netif_dbg(efx, hw, efx->net_dev, "Inserted vf %s rx filter %d\n",
381			  vf->pci_name, rc);
382		vf->rx_filter_id = rc;
383	}
384}
385
386static void __efx_siena_sriov_update_vf_addr(struct siena_vf *vf)
387{
388	struct efx_nic *efx = vf->efx;
389	struct siena_nic_data *nic_data = efx->nic_data;
390
391	efx_siena_sriov_reset_tx_filter(vf);
392	efx_siena_sriov_reset_rx_filter(vf);
393	queue_work(vfdi_workqueue, &nic_data->peer_work);
394}
395
396/* Push the peer list to this VF. The caller must hold status_lock to interlock
397 * with VFDI requests, and they must be serialised against manipulation of
398 * local_page_list, either by acquiring local_lock or by running from
399 * efx_siena_sriov_peer_work()
400 */
401static void __efx_siena_sriov_push_vf_status(struct siena_vf *vf)
402{
403	struct efx_nic *efx = vf->efx;
404	struct siena_nic_data *nic_data = efx->nic_data;
405	struct vfdi_status *status = nic_data->vfdi_status.addr;
406	struct efx_memcpy_req copy[4];
407	struct efx_endpoint_page *epp;
408	unsigned int pos, count;
409	unsigned data_offset;
410	efx_qword_t event;
411
412	WARN_ON(!mutex_is_locked(&vf->status_lock));
413	WARN_ON(!vf->status_addr);
414
415	status->local = vf->addr;
416	status->generation_end = ++status->generation_start;
417
418	memset(copy, '\0', sizeof(copy));
419	/* Write generation_start */
420	copy[0].from_buf = &status->generation_start;
421	copy[0].to_rid = vf->pci_rid;
422	copy[0].to_addr = vf->status_addr + offsetof(struct vfdi_status,
423						     generation_start);
424	copy[0].length = sizeof(status->generation_start);
425	/* DMA the rest of the structure (excluding the generations). This
426	 * assumes that the non-generation portion of vfdi_status is in
427	 * one chunk starting at the version member.
428	 */
429	data_offset = offsetof(struct vfdi_status, version);
430	copy[1].from_rid = efx->pci_dev->devfn;
431	copy[1].from_addr = nic_data->vfdi_status.dma_addr + data_offset;
432	copy[1].to_rid = vf->pci_rid;
433	copy[1].to_addr = vf->status_addr + data_offset;
434	copy[1].length =  status->length - data_offset;
435
436	/* Copy the peer pages */
437	pos = 2;
438	count = 0;
439	list_for_each_entry(epp, &nic_data->local_page_list, link) {
440		if (count == vf->peer_page_count) {
441			/* The VF driver will know they need to provide more
442			 * pages because peer_addr_count is too large.
443			 */
444			break;
445		}
446		copy[pos].from_buf = NULL;
447		copy[pos].from_rid = efx->pci_dev->devfn;
448		copy[pos].from_addr = epp->addr;
449		copy[pos].to_rid = vf->pci_rid;
450		copy[pos].to_addr = vf->peer_page_addrs[count];
451		copy[pos].length = EFX_PAGE_SIZE;
452
453		if (++pos == ARRAY_SIZE(copy)) {
454			efx_siena_sriov_memcpy(efx, copy, ARRAY_SIZE(copy));
455			pos = 0;
456		}
457		++count;
458	}
459
460	/* Write generation_end */
461	copy[pos].from_buf = &status->generation_end;
462	copy[pos].to_rid = vf->pci_rid;
463	copy[pos].to_addr = vf->status_addr + offsetof(struct vfdi_status,
464						       generation_end);
465	copy[pos].length = sizeof(status->generation_end);
466	efx_siena_sriov_memcpy(efx, copy, pos + 1);
467
468	/* Notify the guest */
469	EFX_POPULATE_QWORD_3(event,
470			     FSF_AZ_EV_CODE, FSE_CZ_EV_CODE_USER_EV,
471			     VFDI_EV_SEQ, (vf->msg_seqno & 0xff),
472			     VFDI_EV_TYPE, VFDI_EV_TYPE_STATUS);
473	++vf->msg_seqno;
474	efx_farch_generate_event(efx,
475				 EFX_VI_BASE + vf->index * efx_vf_size(efx),
476				 &event);
477}
478
479static void efx_siena_sriov_bufs(struct efx_nic *efx, unsigned offset,
480				 u64 *addr, unsigned count)
481{
482	efx_qword_t buf;
483	unsigned pos;
484
485	for (pos = 0; pos < count; ++pos) {
486		EFX_POPULATE_QWORD_3(buf,
487				     FRF_AZ_BUF_ADR_REGION, 0,
488				     FRF_AZ_BUF_ADR_FBUF,
489				     addr ? addr[pos] >> 12 : 0,
490				     FRF_AZ_BUF_OWNER_ID_FBUF, 0);
491		efx_sram_writeq(efx, efx->membase + FR_BZ_BUF_FULL_TBL,
492				&buf, offset + pos);
493	}
494}
495
496static bool bad_vf_index(struct efx_nic *efx, unsigned index)
497{
498	return index >= efx_vf_size(efx);
499}
500
501static bool bad_buf_count(unsigned buf_count, unsigned max_entry_count)
502{
503	unsigned max_buf_count = max_entry_count *
504		sizeof(efx_qword_t) / EFX_BUF_SIZE;
505
506	return ((buf_count & (buf_count - 1)) || buf_count > max_buf_count);
507}
508
509/* Check that VI specified by per-port index belongs to a VF.
510 * Optionally set VF index and VI index within the VF.
511 */
512static bool map_vi_index(struct efx_nic *efx, unsigned abs_index,
513			 struct siena_vf **vf_out, unsigned *rel_index_out)
514{
515	struct siena_nic_data *nic_data = efx->nic_data;
516	unsigned vf_i;
517
518	if (abs_index < EFX_VI_BASE)
519		return true;
520	vf_i = (abs_index - EFX_VI_BASE) / efx_vf_size(efx);
521	if (vf_i >= efx->vf_init_count)
522		return true;
523
524	if (vf_out)
525		*vf_out = nic_data->vf + vf_i;
526	if (rel_index_out)
527		*rel_index_out = abs_index % efx_vf_size(efx);
528	return false;
529}
530
531static int efx_vfdi_init_evq(struct siena_vf *vf)
532{
533	struct efx_nic *efx = vf->efx;
534	struct vfdi_req *req = vf->buf.addr;
535	unsigned vf_evq = req->u.init_evq.index;
536	unsigned buf_count = req->u.init_evq.buf_count;
537	unsigned abs_evq = abs_index(vf, vf_evq);
538	unsigned buftbl = EFX_BUFTBL_EVQ_BASE(vf, vf_evq);
539	efx_oword_t reg;
540
541	if (bad_vf_index(efx, vf_evq) ||
542	    bad_buf_count(buf_count, EFX_MAX_VF_EVQ_SIZE)) {
543		if (net_ratelimit())
544			netif_err(efx, hw, efx->net_dev,
545				  "ERROR: Invalid INIT_EVQ from %s: evq %d bufs %d\n",
546				  vf->pci_name, vf_evq, buf_count);
547		return VFDI_RC_EINVAL;
548	}
549
550	efx_siena_sriov_bufs(efx, buftbl, req->u.init_evq.addr, buf_count);
551
552	EFX_POPULATE_OWORD_3(reg,
553			     FRF_CZ_TIMER_Q_EN, 1,
554			     FRF_CZ_HOST_NOTIFY_MODE, 0,
555			     FRF_CZ_TIMER_MODE, FFE_CZ_TIMER_MODE_DIS);
556	efx_writeo_table(efx, &reg, FR_BZ_TIMER_TBL, abs_evq);
557	EFX_POPULATE_OWORD_3(reg,
558			     FRF_AZ_EVQ_EN, 1,
559			     FRF_AZ_EVQ_SIZE, __ffs(buf_count),
560			     FRF_AZ_EVQ_BUF_BASE_ID, buftbl);
561	efx_writeo_table(efx, &reg, FR_BZ_EVQ_PTR_TBL, abs_evq);
562
563	if (vf_evq == 0) {
564		memcpy(vf->evq0_addrs, req->u.init_evq.addr,
565		       buf_count * sizeof(u64));
566		vf->evq0_count = buf_count;
567	}
568
569	return VFDI_RC_SUCCESS;
570}
571
572static int efx_vfdi_init_rxq(struct siena_vf *vf)
573{
574	struct efx_nic *efx = vf->efx;
575	struct vfdi_req *req = vf->buf.addr;
576	unsigned vf_rxq = req->u.init_rxq.index;
577	unsigned vf_evq = req->u.init_rxq.evq;
578	unsigned buf_count = req->u.init_rxq.buf_count;
579	unsigned buftbl = EFX_BUFTBL_RXQ_BASE(vf, vf_rxq);
580	unsigned label;
581	efx_oword_t reg;
582
583	if (bad_vf_index(efx, vf_evq) || bad_vf_index(efx, vf_rxq) ||
584	    vf_rxq >= VF_MAX_RX_QUEUES ||
585	    bad_buf_count(buf_count, EFX_MAX_DMAQ_SIZE)) {
586		if (net_ratelimit())
587			netif_err(efx, hw, efx->net_dev,
588				  "ERROR: Invalid INIT_RXQ from %s: rxq %d evq %d "
589				  "buf_count %d\n", vf->pci_name, vf_rxq,
590				  vf_evq, buf_count);
591		return VFDI_RC_EINVAL;
592	}
593	if (__test_and_set_bit(req->u.init_rxq.index, vf->rxq_mask))
594		++vf->rxq_count;
595	efx_siena_sriov_bufs(efx, buftbl, req->u.init_rxq.addr, buf_count);
596
597	label = req->u.init_rxq.label & EFX_FIELD_MASK(FRF_AZ_RX_DESCQ_LABEL);
598	EFX_POPULATE_OWORD_6(reg,
599			     FRF_AZ_RX_DESCQ_BUF_BASE_ID, buftbl,
600			     FRF_AZ_RX_DESCQ_EVQ_ID, abs_index(vf, vf_evq),
601			     FRF_AZ_RX_DESCQ_LABEL, label,
602			     FRF_AZ_RX_DESCQ_SIZE, __ffs(buf_count),
603			     FRF_AZ_RX_DESCQ_JUMBO,
604			     !!(req->u.init_rxq.flags &
605				VFDI_RXQ_FLAG_SCATTER_EN),
606			     FRF_AZ_RX_DESCQ_EN, 1);
607	efx_writeo_table(efx, &reg, FR_BZ_RX_DESC_PTR_TBL,
608			 abs_index(vf, vf_rxq));
609
610	return VFDI_RC_SUCCESS;
611}
612
613static int efx_vfdi_init_txq(struct siena_vf *vf)
614{
615	struct efx_nic *efx = vf->efx;
616	struct vfdi_req *req = vf->buf.addr;
617	unsigned vf_txq = req->u.init_txq.index;
618	unsigned vf_evq = req->u.init_txq.evq;
619	unsigned buf_count = req->u.init_txq.buf_count;
620	unsigned buftbl = EFX_BUFTBL_TXQ_BASE(vf, vf_txq);
621	unsigned label, eth_filt_en;
622	efx_oword_t reg;
623
624	if (bad_vf_index(efx, vf_evq) || bad_vf_index(efx, vf_txq) ||
625	    vf_txq >= vf_max_tx_channels ||
626	    bad_buf_count(buf_count, EFX_MAX_DMAQ_SIZE)) {
627		if (net_ratelimit())
628			netif_err(efx, hw, efx->net_dev,
629				  "ERROR: Invalid INIT_TXQ from %s: txq %d evq %d "
630				  "buf_count %d\n", vf->pci_name, vf_txq,
631				  vf_evq, buf_count);
632		return VFDI_RC_EINVAL;
633	}
634
635	mutex_lock(&vf->txq_lock);
636	if (__test_and_set_bit(req->u.init_txq.index, vf->txq_mask))
637		++vf->txq_count;
638	mutex_unlock(&vf->txq_lock);
639	efx_siena_sriov_bufs(efx, buftbl, req->u.init_txq.addr, buf_count);
640
641	eth_filt_en = vf->tx_filter_mode == VF_TX_FILTER_ON;
642
643	label = req->u.init_txq.label & EFX_FIELD_MASK(FRF_AZ_TX_DESCQ_LABEL);
644	EFX_POPULATE_OWORD_8(reg,
645			     FRF_CZ_TX_DPT_Q_MASK_WIDTH, min(efx->vi_scale, 1U),
646			     FRF_CZ_TX_DPT_ETH_FILT_EN, eth_filt_en,
647			     FRF_AZ_TX_DESCQ_EN, 1,
648			     FRF_AZ_TX_DESCQ_BUF_BASE_ID, buftbl,
649			     FRF_AZ_TX_DESCQ_EVQ_ID, abs_index(vf, vf_evq),
650			     FRF_AZ_TX_DESCQ_LABEL, label,
651			     FRF_AZ_TX_DESCQ_SIZE, __ffs(buf_count),
652			     FRF_BZ_TX_NON_IP_DROP_DIS, 1);
653	efx_writeo_table(efx, &reg, FR_BZ_TX_DESC_PTR_TBL,
654			 abs_index(vf, vf_txq));
655
656	return VFDI_RC_SUCCESS;
657}
658
659/* Returns true when efx_vfdi_fini_all_queues should wake */
660static bool efx_vfdi_flush_wake(struct siena_vf *vf)
661{
662	/* Ensure that all updates are visible to efx_vfdi_fini_all_queues() */
663	smp_mb();
664
665	return (!vf->txq_count && !vf->rxq_count) ||
666		atomic_read(&vf->rxq_retry_count);
667}
668
669static void efx_vfdi_flush_clear(struct siena_vf *vf)
670{
671	memset(vf->txq_mask, 0, sizeof(vf->txq_mask));
672	vf->txq_count = 0;
673	memset(vf->rxq_mask, 0, sizeof(vf->rxq_mask));
674	vf->rxq_count = 0;
675	memset(vf->rxq_retry_mask, 0, sizeof(vf->rxq_retry_mask));
676	atomic_set(&vf->rxq_retry_count, 0);
677}
678
679static int efx_vfdi_fini_all_queues(struct siena_vf *vf)
680{
681	struct efx_nic *efx = vf->efx;
682	efx_oword_t reg;
683	unsigned count = efx_vf_size(efx);
684	unsigned vf_offset = EFX_VI_BASE + vf->index * efx_vf_size(efx);
685	unsigned timeout = HZ;
686	unsigned index, rxqs_count;
687	MCDI_DECLARE_BUF(inbuf, MC_CMD_FLUSH_RX_QUEUES_IN_LENMAX);
688	int rc;
689
690	BUILD_BUG_ON(VF_MAX_RX_QUEUES >
691		     MC_CMD_FLUSH_RX_QUEUES_IN_QID_OFST_MAXNUM);
692
693	rtnl_lock();
694	siena_prepare_flush(efx);
695	rtnl_unlock();
696
697	/* Flush all the initialized queues */
698	rxqs_count = 0;
699	for (index = 0; index < count; ++index) {
700		if (test_bit(index, vf->txq_mask)) {
701			EFX_POPULATE_OWORD_2(reg,
702					     FRF_AZ_TX_FLUSH_DESCQ_CMD, 1,
703					     FRF_AZ_TX_FLUSH_DESCQ,
704					     vf_offset + index);
705			efx_writeo(efx, &reg, FR_AZ_TX_FLUSH_DESCQ);
706		}
707		if (test_bit(index, vf->rxq_mask)) {
708			MCDI_SET_ARRAY_DWORD(
709				inbuf, FLUSH_RX_QUEUES_IN_QID_OFST,
710				rxqs_count, vf_offset + index);
711			rxqs_count++;
712		}
713	}
714
715	atomic_set(&vf->rxq_retry_count, 0);
716	while (timeout && (vf->rxq_count || vf->txq_count)) {
717		rc = efx_mcdi_rpc(efx, MC_CMD_FLUSH_RX_QUEUES, inbuf,
718				  MC_CMD_FLUSH_RX_QUEUES_IN_LEN(rxqs_count),
719				  NULL, 0, NULL);
720		WARN_ON(rc < 0);
721
722		timeout = wait_event_timeout(vf->flush_waitq,
723					     efx_vfdi_flush_wake(vf),
724					     timeout);
725		rxqs_count = 0;
726		for (index = 0; index < count; ++index) {
727			if (test_and_clear_bit(index, vf->rxq_retry_mask)) {
728				atomic_dec(&vf->rxq_retry_count);
729				MCDI_SET_ARRAY_DWORD(
730					inbuf, FLUSH_RX_QUEUES_IN_QID_OFST,
731					rxqs_count, vf_offset + index);
732				rxqs_count++;
733			}
734		}
735	}
736
737	rtnl_lock();
738	siena_finish_flush(efx);
739	rtnl_unlock();
740
741	/* Irrespective of success/failure, fini the queues */
742	EFX_ZERO_OWORD(reg);
743	for (index = 0; index < count; ++index) {
744		efx_writeo_table(efx, &reg, FR_BZ_RX_DESC_PTR_TBL,
745				 vf_offset + index);
746		efx_writeo_table(efx, &reg, FR_BZ_TX_DESC_PTR_TBL,
747				 vf_offset + index);
748		efx_writeo_table(efx, &reg, FR_BZ_EVQ_PTR_TBL,
749				 vf_offset + index);
750		efx_writeo_table(efx, &reg, FR_BZ_TIMER_TBL,
751				 vf_offset + index);
752	}
753	efx_siena_sriov_bufs(efx, vf->buftbl_base, NULL,
754			     EFX_VF_BUFTBL_PER_VI * efx_vf_size(efx));
755	efx_vfdi_flush_clear(vf);
756
757	vf->evq0_count = 0;
758
759	return timeout ? 0 : VFDI_RC_ETIMEDOUT;
760}
761
762static int efx_vfdi_insert_filter(struct siena_vf *vf)
763{
764	struct efx_nic *efx = vf->efx;
765	struct siena_nic_data *nic_data = efx->nic_data;
766	struct vfdi_req *req = vf->buf.addr;
767	unsigned vf_rxq = req->u.mac_filter.rxq;
768	unsigned flags;
769
770	if (bad_vf_index(efx, vf_rxq) || vf->rx_filtering) {
771		if (net_ratelimit())
772			netif_err(efx, hw, efx->net_dev,
773				  "ERROR: Invalid INSERT_FILTER from %s: rxq %d "
774				  "flags 0x%x\n", vf->pci_name, vf_rxq,
775				  req->u.mac_filter.flags);
776		return VFDI_RC_EINVAL;
777	}
778
779	flags = 0;
780	if (req->u.mac_filter.flags & VFDI_MAC_FILTER_FLAG_RSS)
781		flags |= EFX_FILTER_FLAG_RX_RSS;
782	if (req->u.mac_filter.flags & VFDI_MAC_FILTER_FLAG_SCATTER)
783		flags |= EFX_FILTER_FLAG_RX_SCATTER;
784	vf->rx_filter_flags = flags;
785	vf->rx_filter_qid = vf_rxq;
786	vf->rx_filtering = true;
787
788	efx_siena_sriov_reset_rx_filter(vf);
789	queue_work(vfdi_workqueue, &nic_data->peer_work);
790
791	return VFDI_RC_SUCCESS;
792}
793
794static int efx_vfdi_remove_all_filters(struct siena_vf *vf)
795{
796	struct efx_nic *efx = vf->efx;
797	struct siena_nic_data *nic_data = efx->nic_data;
798
799	vf->rx_filtering = false;
800	efx_siena_sriov_reset_rx_filter(vf);
801	queue_work(vfdi_workqueue, &nic_data->peer_work);
802
803	return VFDI_RC_SUCCESS;
804}
805
806static int efx_vfdi_set_status_page(struct siena_vf *vf)
807{
808	struct efx_nic *efx = vf->efx;
809	struct siena_nic_data *nic_data = efx->nic_data;
810	struct vfdi_req *req = vf->buf.addr;
811	u64 page_count = req->u.set_status_page.peer_page_count;
812	u64 max_page_count =
813		(EFX_PAGE_SIZE -
814		 offsetof(struct vfdi_req, u.set_status_page.peer_page_addr[0]))
815		/ sizeof(req->u.set_status_page.peer_page_addr[0]);
816
817	if (!req->u.set_status_page.dma_addr || page_count > max_page_count) {
818		if (net_ratelimit())
819			netif_err(efx, hw, efx->net_dev,
820				  "ERROR: Invalid SET_STATUS_PAGE from %s\n",
821				  vf->pci_name);
822		return VFDI_RC_EINVAL;
823	}
824
825	mutex_lock(&nic_data->local_lock);
826	mutex_lock(&vf->status_lock);
827	vf->status_addr = req->u.set_status_page.dma_addr;
828
829	kfree(vf->peer_page_addrs);
830	vf->peer_page_addrs = NULL;
831	vf->peer_page_count = 0;
832
833	if (page_count) {
834		vf->peer_page_addrs = kcalloc(page_count, sizeof(u64),
835					      GFP_KERNEL);
836		if (vf->peer_page_addrs) {
837			memcpy(vf->peer_page_addrs,
838			       req->u.set_status_page.peer_page_addr,
839			       page_count * sizeof(u64));
840			vf->peer_page_count = page_count;
841		}
842	}
843
844	__efx_siena_sriov_push_vf_status(vf);
845	mutex_unlock(&vf->status_lock);
846	mutex_unlock(&nic_data->local_lock);
847
848	return VFDI_RC_SUCCESS;
849}
850
851static int efx_vfdi_clear_status_page(struct siena_vf *vf)
852{
853	mutex_lock(&vf->status_lock);
854	vf->status_addr = 0;
855	mutex_unlock(&vf->status_lock);
856
857	return VFDI_RC_SUCCESS;
858}
859
860typedef int (*efx_vfdi_op_t)(struct siena_vf *vf);
861
862static const efx_vfdi_op_t vfdi_ops[VFDI_OP_LIMIT] = {
863	[VFDI_OP_INIT_EVQ] = efx_vfdi_init_evq,
864	[VFDI_OP_INIT_TXQ] = efx_vfdi_init_txq,
865	[VFDI_OP_INIT_RXQ] = efx_vfdi_init_rxq,
866	[VFDI_OP_FINI_ALL_QUEUES] = efx_vfdi_fini_all_queues,
867	[VFDI_OP_INSERT_FILTER] = efx_vfdi_insert_filter,
868	[VFDI_OP_REMOVE_ALL_FILTERS] = efx_vfdi_remove_all_filters,
869	[VFDI_OP_SET_STATUS_PAGE] = efx_vfdi_set_status_page,
870	[VFDI_OP_CLEAR_STATUS_PAGE] = efx_vfdi_clear_status_page,
871};
872
873static void efx_siena_sriov_vfdi(struct work_struct *work)
874{
875	struct siena_vf *vf = container_of(work, struct siena_vf, req);
876	struct efx_nic *efx = vf->efx;
877	struct vfdi_req *req = vf->buf.addr;
878	struct efx_memcpy_req copy[2];
879	int rc;
880
881	/* Copy this page into the local address space */
882	memset(copy, '\0', sizeof(copy));
883	copy[0].from_rid = vf->pci_rid;
884	copy[0].from_addr = vf->req_addr;
885	copy[0].to_rid = efx->pci_dev->devfn;
886	copy[0].to_addr = vf->buf.dma_addr;
887	copy[0].length = EFX_PAGE_SIZE;
888	rc = efx_siena_sriov_memcpy(efx, copy, 1);
889	if (rc) {
890		/* If we can't get the request, we can't reply to the caller */
891		if (net_ratelimit())
892			netif_err(efx, hw, efx->net_dev,
893				  "ERROR: Unable to fetch VFDI request from %s rc %d\n",
894				  vf->pci_name, -rc);
895		vf->busy = false;
896		return;
897	}
898
899	if (req->op < VFDI_OP_LIMIT && vfdi_ops[req->op] != NULL) {
900		rc = vfdi_ops[req->op](vf);
901		if (rc == 0) {
902			netif_dbg(efx, hw, efx->net_dev,
903				  "vfdi request %d from %s ok\n",
904				  req->op, vf->pci_name);
905		}
906	} else {
907		netif_dbg(efx, hw, efx->net_dev,
908			  "ERROR: Unrecognised request %d from VF %s addr "
909			  "%llx\n", req->op, vf->pci_name,
910			  (unsigned long long)vf->req_addr);
911		rc = VFDI_RC_EOPNOTSUPP;
912	}
913
914	/* Allow subsequent VF requests */
915	vf->busy = false;
916	smp_wmb();
917
918	/* Respond to the request */
919	req->rc = rc;
920	req->op = VFDI_OP_RESPONSE;
921
922	memset(copy, '\0', sizeof(copy));
923	copy[0].from_buf = &req->rc;
924	copy[0].to_rid = vf->pci_rid;
925	copy[0].to_addr = vf->req_addr + offsetof(struct vfdi_req, rc);
926	copy[0].length = sizeof(req->rc);
927	copy[1].from_buf = &req->op;
928	copy[1].to_rid = vf->pci_rid;
929	copy[1].to_addr = vf->req_addr + offsetof(struct vfdi_req, op);
930	copy[1].length = sizeof(req->op);
931
932	(void)efx_siena_sriov_memcpy(efx, copy, ARRAY_SIZE(copy));
933}
934
935
936
937/* After a reset the event queues inside the guests no longer exist. Fill the
938 * event ring in guest memory with VFDI reset events, then (re-initialise) the
939 * event queue to raise an interrupt. The guest driver will then recover.
940 */
941
942static void efx_siena_sriov_reset_vf(struct siena_vf *vf,
943				     struct efx_buffer *buffer)
944{
945	struct efx_nic *efx = vf->efx;
946	struct efx_memcpy_req copy_req[4];
947	efx_qword_t event;
948	unsigned int pos, count, k, buftbl, abs_evq;
949	efx_oword_t reg;
950	efx_dword_t ptr;
951	int rc;
952
953	BUG_ON(buffer->len != EFX_PAGE_SIZE);
954
955	if (!vf->evq0_count)
956		return;
957	BUG_ON(vf->evq0_count & (vf->evq0_count - 1));
958
959	mutex_lock(&vf->status_lock);
960	EFX_POPULATE_QWORD_3(event,
961			     FSF_AZ_EV_CODE, FSE_CZ_EV_CODE_USER_EV,
962			     VFDI_EV_SEQ, vf->msg_seqno,
963			     VFDI_EV_TYPE, VFDI_EV_TYPE_RESET);
964	vf->msg_seqno++;
965	for (pos = 0; pos < EFX_PAGE_SIZE; pos += sizeof(event))
966		memcpy(buffer->addr + pos, &event, sizeof(event));
967
968	for (pos = 0; pos < vf->evq0_count; pos += count) {
969		count = min_t(unsigned, vf->evq0_count - pos,
970			      ARRAY_SIZE(copy_req));
971		for (k = 0; k < count; k++) {
972			copy_req[k].from_buf = NULL;
973			copy_req[k].from_rid = efx->pci_dev->devfn;
974			copy_req[k].from_addr = buffer->dma_addr;
975			copy_req[k].to_rid = vf->pci_rid;
976			copy_req[k].to_addr = vf->evq0_addrs[pos + k];
977			copy_req[k].length = EFX_PAGE_SIZE;
978		}
979		rc = efx_siena_sriov_memcpy(efx, copy_req, count);
980		if (rc) {
981			if (net_ratelimit())
982				netif_err(efx, hw, efx->net_dev,
983					  "ERROR: Unable to notify %s of reset"
984					  ": %d\n", vf->pci_name, -rc);
985			break;
986		}
987	}
988
989	/* Reinitialise, arm and trigger evq0 */
990	abs_evq = abs_index(vf, 0);
991	buftbl = EFX_BUFTBL_EVQ_BASE(vf, 0);
992	efx_siena_sriov_bufs(efx, buftbl, vf->evq0_addrs, vf->evq0_count);
993
994	EFX_POPULATE_OWORD_3(reg,
995			     FRF_CZ_TIMER_Q_EN, 1,
996			     FRF_CZ_HOST_NOTIFY_MODE, 0,
997			     FRF_CZ_TIMER_MODE, FFE_CZ_TIMER_MODE_DIS);
998	efx_writeo_table(efx, &reg, FR_BZ_TIMER_TBL, abs_evq);
999	EFX_POPULATE_OWORD_3(reg,
1000			     FRF_AZ_EVQ_EN, 1,
1001			     FRF_AZ_EVQ_SIZE, __ffs(vf->evq0_count),
1002			     FRF_AZ_EVQ_BUF_BASE_ID, buftbl);
1003	efx_writeo_table(efx, &reg, FR_BZ_EVQ_PTR_TBL, abs_evq);
1004	EFX_POPULATE_DWORD_1(ptr, FRF_AZ_EVQ_RPTR, 0);
1005	efx_writed(efx, &ptr, FR_BZ_EVQ_RPTR + FR_BZ_EVQ_RPTR_STEP * abs_evq);
1006
1007	mutex_unlock(&vf->status_lock);
1008}
1009
1010static void efx_siena_sriov_reset_vf_work(struct work_struct *work)
1011{
1012	struct siena_vf *vf = container_of(work, struct siena_vf, req);
1013	struct efx_nic *efx = vf->efx;
1014	struct efx_buffer buf;
1015
1016	if (!efx_nic_alloc_buffer(efx, &buf, EFX_PAGE_SIZE, GFP_NOIO)) {
1017		efx_siena_sriov_reset_vf(vf, &buf);
1018		efx_nic_free_buffer(efx, &buf);
1019	}
1020}
1021
1022static void efx_siena_sriov_handle_no_channel(struct efx_nic *efx)
1023{
1024	netif_err(efx, drv, efx->net_dev,
1025		  "ERROR: IOV requires MSI-X and 1 additional interrupt"
1026		  "vector. IOV disabled\n");
1027	efx->vf_count = 0;
1028}
1029
1030static int efx_siena_sriov_probe_channel(struct efx_channel *channel)
1031{
1032	struct siena_nic_data *nic_data = channel->efx->nic_data;
1033	nic_data->vfdi_channel = channel;
1034
1035	return 0;
1036}
1037
1038static void
1039efx_siena_sriov_get_channel_name(struct efx_channel *channel,
1040				 char *buf, size_t len)
1041{
1042	snprintf(buf, len, "%s-iov", channel->efx->name);
1043}
1044
1045static const struct efx_channel_type efx_siena_sriov_channel_type = {
1046	.handle_no_channel	= efx_siena_sriov_handle_no_channel,
1047	.pre_probe		= efx_siena_sriov_probe_channel,
1048	.post_remove		= efx_channel_dummy_op_void,
1049	.get_name		= efx_siena_sriov_get_channel_name,
1050	/* no copy operation; channel must not be reallocated */
1051	.keep_eventq		= true,
1052};
1053
1054void efx_siena_sriov_probe(struct efx_nic *efx)
1055{
1056	unsigned count;
1057
1058	if (!max_vfs)
1059		return;
1060
1061	if (efx_siena_sriov_cmd(efx, false, &efx->vi_scale, &count)) {
1062		netif_info(efx, probe, efx->net_dev, "no SR-IOV VFs probed\n");
1063		return;
1064	}
1065	if (count > 0 && count > max_vfs)
1066		count = max_vfs;
1067
1068	/* efx_nic_dimension_resources() will reduce vf_count as appopriate */
1069	efx->vf_count = count;
1070
1071	efx->extra_channel_type[EFX_EXTRA_CHANNEL_IOV] = &efx_siena_sriov_channel_type;
1072}
1073
1074/* Copy the list of individual addresses into the vfdi_status.peers
1075 * array and auxiliary pages, protected by %local_lock. Drop that lock
1076 * and then broadcast the address list to every VF.
1077 */
1078static void efx_siena_sriov_peer_work(struct work_struct *data)
1079{
1080	struct siena_nic_data *nic_data = container_of(data,
1081						       struct siena_nic_data,
1082						       peer_work);
1083	struct efx_nic *efx = nic_data->efx;
1084	struct vfdi_status *vfdi_status = nic_data->vfdi_status.addr;
1085	struct siena_vf *vf;
1086	struct efx_local_addr *local_addr;
1087	struct vfdi_endpoint *peer;
1088	struct efx_endpoint_page *epp;
1089	struct list_head pages;
1090	unsigned int peer_space;
1091	unsigned int peer_count;
1092	unsigned int pos;
1093
1094	mutex_lock(&nic_data->local_lock);
1095
1096	/* Move the existing peer pages off %local_page_list */
1097	INIT_LIST_HEAD(&pages);
1098	list_splice_tail_init(&nic_data->local_page_list, &pages);
1099
1100	/* Populate the VF addresses starting from entry 1 (entry 0 is
1101	 * the PF address)
1102	 */
1103	peer = vfdi_status->peers + 1;
1104	peer_space = ARRAY_SIZE(vfdi_status->peers) - 1;
1105	peer_count = 1;
1106	for (pos = 0; pos < efx->vf_count; ++pos) {
1107		vf = nic_data->vf + pos;
1108
1109		mutex_lock(&vf->status_lock);
1110		if (vf->rx_filtering && !is_zero_ether_addr(vf->addr.mac_addr)) {
1111			*peer++ = vf->addr;
1112			++peer_count;
1113			--peer_space;
1114			BUG_ON(peer_space == 0);
1115		}
1116		mutex_unlock(&vf->status_lock);
1117	}
1118
1119	/* Fill the remaining addresses */
1120	list_for_each_entry(local_addr, &nic_data->local_addr_list, link) {
1121		ether_addr_copy(peer->mac_addr, local_addr->addr);
1122		peer->tci = 0;
1123		++peer;
1124		++peer_count;
1125		if (--peer_space == 0) {
1126			if (list_empty(&pages)) {
1127				epp = kmalloc(sizeof(*epp), GFP_KERNEL);
1128				if (!epp)
1129					break;
1130				epp->ptr = dma_alloc_coherent(
1131					&efx->pci_dev->dev, EFX_PAGE_SIZE,
1132					&epp->addr, GFP_KERNEL);
1133				if (!epp->ptr) {
1134					kfree(epp);
1135					break;
1136				}
1137			} else {
1138				epp = list_first_entry(
1139					&pages, struct efx_endpoint_page, link);
1140				list_del(&epp->link);
1141			}
1142
1143			list_add_tail(&epp->link, &nic_data->local_page_list);
1144			peer = (struct vfdi_endpoint *)epp->ptr;
1145			peer_space = EFX_PAGE_SIZE / sizeof(struct vfdi_endpoint);
1146		}
1147	}
1148	vfdi_status->peer_count = peer_count;
1149	mutex_unlock(&nic_data->local_lock);
1150
1151	/* Free any now unused endpoint pages */
1152	while (!list_empty(&pages)) {
1153		epp = list_first_entry(
1154			&pages, struct efx_endpoint_page, link);
1155		list_del(&epp->link);
1156		dma_free_coherent(&efx->pci_dev->dev, EFX_PAGE_SIZE,
1157				  epp->ptr, epp->addr);
1158		kfree(epp);
1159	}
1160
1161	/* Finally, push the pages */
1162	for (pos = 0; pos < efx->vf_count; ++pos) {
1163		vf = nic_data->vf + pos;
1164
1165		mutex_lock(&vf->status_lock);
1166		if (vf->status_addr)
1167			__efx_siena_sriov_push_vf_status(vf);
1168		mutex_unlock(&vf->status_lock);
1169	}
1170}
1171
1172static void efx_siena_sriov_free_local(struct efx_nic *efx)
1173{
1174	struct siena_nic_data *nic_data = efx->nic_data;
1175	struct efx_local_addr *local_addr;
1176	struct efx_endpoint_page *epp;
1177
1178	while (!list_empty(&nic_data->local_addr_list)) {
1179		local_addr = list_first_entry(&nic_data->local_addr_list,
1180					      struct efx_local_addr, link);
1181		list_del(&local_addr->link);
1182		kfree(local_addr);
1183	}
1184
1185	while (!list_empty(&nic_data->local_page_list)) {
1186		epp = list_first_entry(&nic_data->local_page_list,
1187				       struct efx_endpoint_page, link);
1188		list_del(&epp->link);
1189		dma_free_coherent(&efx->pci_dev->dev, EFX_PAGE_SIZE,
1190				  epp->ptr, epp->addr);
1191		kfree(epp);
1192	}
1193}
1194
1195static int efx_siena_sriov_vf_alloc(struct efx_nic *efx)
1196{
1197	unsigned index;
1198	struct siena_vf *vf;
1199	struct siena_nic_data *nic_data = efx->nic_data;
1200
1201	nic_data->vf = kcalloc(efx->vf_count, sizeof(*nic_data->vf),
1202			       GFP_KERNEL);
1203	if (!nic_data->vf)
1204		return -ENOMEM;
1205
1206	for (index = 0; index < efx->vf_count; ++index) {
1207		vf = nic_data->vf + index;
1208
1209		vf->efx = efx;
1210		vf->index = index;
1211		vf->rx_filter_id = -1;
1212		vf->tx_filter_mode = VF_TX_FILTER_AUTO;
1213		vf->tx_filter_id = -1;
1214		INIT_WORK(&vf->req, efx_siena_sriov_vfdi);
1215		INIT_WORK(&vf->reset_work, efx_siena_sriov_reset_vf_work);
1216		init_waitqueue_head(&vf->flush_waitq);
1217		mutex_init(&vf->status_lock);
1218		mutex_init(&vf->txq_lock);
1219	}
1220
1221	return 0;
1222}
1223
1224static void efx_siena_sriov_vfs_fini(struct efx_nic *efx)
1225{
1226	struct siena_nic_data *nic_data = efx->nic_data;
1227	struct siena_vf *vf;
1228	unsigned int pos;
1229
1230	for (pos = 0; pos < efx->vf_count; ++pos) {
1231		vf = nic_data->vf + pos;
1232
1233		efx_nic_free_buffer(efx, &vf->buf);
1234		kfree(vf->peer_page_addrs);
1235		vf->peer_page_addrs = NULL;
1236		vf->peer_page_count = 0;
1237
1238		vf->evq0_count = 0;
1239	}
1240}
1241
1242static int efx_siena_sriov_vfs_init(struct efx_nic *efx)
1243{
1244	struct pci_dev *pci_dev = efx->pci_dev;
1245	struct siena_nic_data *nic_data = efx->nic_data;
1246	unsigned index, devfn, sriov, buftbl_base;
1247	u16 offset, stride;
1248	struct siena_vf *vf;
1249	int rc;
1250
1251	sriov = pci_find_ext_capability(pci_dev, PCI_EXT_CAP_ID_SRIOV);
1252	if (!sriov)
1253		return -ENOENT;
1254
1255	pci_read_config_word(pci_dev, sriov + PCI_SRIOV_VF_OFFSET, &offset);
1256	pci_read_config_word(pci_dev, sriov + PCI_SRIOV_VF_STRIDE, &stride);
1257
1258	buftbl_base = nic_data->vf_buftbl_base;
1259	devfn = pci_dev->devfn + offset;
1260	for (index = 0; index < efx->vf_count; ++index) {
1261		vf = nic_data->vf + index;
1262
1263		/* Reserve buffer entries */
1264		vf->buftbl_base = buftbl_base;
1265		buftbl_base += EFX_VF_BUFTBL_PER_VI * efx_vf_size(efx);
1266
1267		vf->pci_rid = devfn;
1268		snprintf(vf->pci_name, sizeof(vf->pci_name),
1269			 "%04x:%02x:%02x.%d",
1270			 pci_domain_nr(pci_dev->bus), pci_dev->bus->number,
1271			 PCI_SLOT(devfn), PCI_FUNC(devfn));
1272
1273		rc = efx_nic_alloc_buffer(efx, &vf->buf, EFX_PAGE_SIZE,
1274					  GFP_KERNEL);
1275		if (rc)
1276			goto fail;
1277
1278		devfn += stride;
1279	}
1280
1281	return 0;
1282
1283fail:
1284	efx_siena_sriov_vfs_fini(efx);
1285	return rc;
1286}
1287
1288int efx_siena_sriov_init(struct efx_nic *efx)
1289{
1290	struct net_device *net_dev = efx->net_dev;
1291	struct siena_nic_data *nic_data = efx->nic_data;
1292	struct vfdi_status *vfdi_status;
1293	int rc;
1294
1295	/* Ensure there's room for vf_channel */
1296	BUILD_BUG_ON(EFX_MAX_CHANNELS + 1 >= EFX_VI_BASE);
1297	/* Ensure that VI_BASE is aligned on VI_SCALE */
1298	BUILD_BUG_ON(EFX_VI_BASE & ((1 << EFX_VI_SCALE_MAX) - 1));
1299
1300	if (efx->vf_count == 0)
1301		return 0;
1302
1303	rc = efx_siena_sriov_cmd(efx, true, NULL, NULL);
1304	if (rc)
1305		goto fail_cmd;
1306
1307	rc = efx_nic_alloc_buffer(efx, &nic_data->vfdi_status,
1308				  sizeof(*vfdi_status), GFP_KERNEL);
1309	if (rc)
1310		goto fail_status;
1311	vfdi_status = nic_data->vfdi_status.addr;
1312	memset(vfdi_status, 0, sizeof(*vfdi_status));
1313	vfdi_status->version = 1;
1314	vfdi_status->length = sizeof(*vfdi_status);
1315	vfdi_status->max_tx_channels = vf_max_tx_channels;
1316	vfdi_status->vi_scale = efx->vi_scale;
1317	vfdi_status->rss_rxq_count = efx->rss_spread;
1318	vfdi_status->peer_count = 1 + efx->vf_count;
1319	vfdi_status->timer_quantum_ns = efx->timer_quantum_ns;
1320
1321	rc = efx_siena_sriov_vf_alloc(efx);
1322	if (rc)
1323		goto fail_alloc;
1324
1325	mutex_init(&nic_data->local_lock);
1326	INIT_WORK(&nic_data->peer_work, efx_siena_sriov_peer_work);
1327	INIT_LIST_HEAD(&nic_data->local_addr_list);
1328	INIT_LIST_HEAD(&nic_data->local_page_list);
1329
1330	rc = efx_siena_sriov_vfs_init(efx);
1331	if (rc)
1332		goto fail_vfs;
1333
1334	rtnl_lock();
1335	ether_addr_copy(vfdi_status->peers[0].mac_addr, net_dev->dev_addr);
1336	efx->vf_init_count = efx->vf_count;
1337	rtnl_unlock();
1338
1339	efx_siena_sriov_usrev(efx, true);
1340
1341	/* At this point we must be ready to accept VFDI requests */
1342
1343	rc = pci_enable_sriov(efx->pci_dev, efx->vf_count);
1344	if (rc)
1345		goto fail_pci;
1346
1347	netif_info(efx, probe, net_dev,
1348		   "enabled SR-IOV for %d VFs, %d VI per VF\n",
1349		   efx->vf_count, efx_vf_size(efx));
1350	return 0;
1351
1352fail_pci:
1353	efx_siena_sriov_usrev(efx, false);
1354	rtnl_lock();
1355	efx->vf_init_count = 0;
1356	rtnl_unlock();
1357	efx_siena_sriov_vfs_fini(efx);
1358fail_vfs:
1359	cancel_work_sync(&nic_data->peer_work);
1360	efx_siena_sriov_free_local(efx);
1361	kfree(nic_data->vf);
1362fail_alloc:
1363	efx_nic_free_buffer(efx, &nic_data->vfdi_status);
1364fail_status:
1365	efx_siena_sriov_cmd(efx, false, NULL, NULL);
1366fail_cmd:
1367	return rc;
1368}
1369
1370void efx_siena_sriov_fini(struct efx_nic *efx)
1371{
1372	struct siena_vf *vf;
1373	unsigned int pos;
1374	struct siena_nic_data *nic_data = efx->nic_data;
1375
1376	if (efx->vf_init_count == 0)
1377		return;
1378
1379	/* Disable all interfaces to reconfiguration */
1380	BUG_ON(nic_data->vfdi_channel->enabled);
1381	efx_siena_sriov_usrev(efx, false);
1382	rtnl_lock();
1383	efx->vf_init_count = 0;
1384	rtnl_unlock();
1385
1386	/* Flush all reconfiguration work */
1387	for (pos = 0; pos < efx->vf_count; ++pos) {
1388		vf = nic_data->vf + pos;
1389		cancel_work_sync(&vf->req);
1390		cancel_work_sync(&vf->reset_work);
1391	}
1392	cancel_work_sync(&nic_data->peer_work);
1393
1394	pci_disable_sriov(efx->pci_dev);
1395
1396	/* Tear down back-end state */
1397	efx_siena_sriov_vfs_fini(efx);
1398	efx_siena_sriov_free_local(efx);
1399	kfree(nic_data->vf);
1400	efx_nic_free_buffer(efx, &nic_data->vfdi_status);
1401	efx_siena_sriov_cmd(efx, false, NULL, NULL);
1402}
1403
1404void efx_siena_sriov_event(struct efx_channel *channel, efx_qword_t *event)
1405{
1406	struct efx_nic *efx = channel->efx;
1407	struct siena_vf *vf;
1408	unsigned qid, seq, type, data;
1409
1410	qid = EFX_QWORD_FIELD(*event, FSF_CZ_USER_QID);
1411
1412	/* USR_EV_REG_VALUE is dword0, so access the VFDI_EV fields directly */
1413	BUILD_BUG_ON(FSF_CZ_USER_EV_REG_VALUE_LBN != 0);
1414	seq = EFX_QWORD_FIELD(*event, VFDI_EV_SEQ);
1415	type = EFX_QWORD_FIELD(*event, VFDI_EV_TYPE);
1416	data = EFX_QWORD_FIELD(*event, VFDI_EV_DATA);
1417
1418	netif_vdbg(efx, hw, efx->net_dev,
1419		   "USR_EV event from qid %d seq 0x%x type %d data 0x%x\n",
1420		   qid, seq, type, data);
1421
1422	if (map_vi_index(efx, qid, &vf, NULL))
1423		return;
1424	if (vf->busy)
1425		goto error;
1426
1427	if (type == VFDI_EV_TYPE_REQ_WORD0) {
1428		/* Resynchronise */
1429		vf->req_type = VFDI_EV_TYPE_REQ_WORD0;
1430		vf->req_seqno = seq + 1;
1431		vf->req_addr = 0;
1432	} else if (seq != (vf->req_seqno++ & 0xff) || type != vf->req_type)
1433		goto error;
1434
1435	switch (vf->req_type) {
1436	case VFDI_EV_TYPE_REQ_WORD0:
1437	case VFDI_EV_TYPE_REQ_WORD1:
1438	case VFDI_EV_TYPE_REQ_WORD2:
1439		vf->req_addr |= (u64)data << (vf->req_type << 4);
1440		++vf->req_type;
1441		return;
1442
1443	case VFDI_EV_TYPE_REQ_WORD3:
1444		vf->req_addr |= (u64)data << 48;
1445		vf->req_type = VFDI_EV_TYPE_REQ_WORD0;
1446		vf->busy = true;
1447		queue_work(vfdi_workqueue, &vf->req);
1448		return;
1449	}
1450
1451error:
1452	if (net_ratelimit())
1453		netif_err(efx, hw, efx->net_dev,
1454			  "ERROR: Screaming VFDI request from %s\n",
1455			  vf->pci_name);
1456	/* Reset the request and sequence number */
1457	vf->req_type = VFDI_EV_TYPE_REQ_WORD0;
1458	vf->req_seqno = seq + 1;
1459}
1460
1461void efx_siena_sriov_flr(struct efx_nic *efx, unsigned vf_i)
1462{
1463	struct siena_nic_data *nic_data = efx->nic_data;
1464	struct siena_vf *vf;
1465
1466	if (vf_i > efx->vf_init_count)
1467		return;
1468	vf = nic_data->vf + vf_i;
1469	netif_info(efx, hw, efx->net_dev,
1470		   "FLR on VF %s\n", vf->pci_name);
1471
1472	vf->status_addr = 0;
1473	efx_vfdi_remove_all_filters(vf);
1474	efx_vfdi_flush_clear(vf);
1475
1476	vf->evq0_count = 0;
1477}
1478
1479int efx_siena_sriov_mac_address_changed(struct efx_nic *efx)
1480{
1481	struct siena_nic_data *nic_data = efx->nic_data;
1482	struct vfdi_status *vfdi_status = nic_data->vfdi_status.addr;
1483
1484	if (!efx->vf_init_count)
1485		return 0;
1486	ether_addr_copy(vfdi_status->peers[0].mac_addr,
1487			efx->net_dev->dev_addr);
1488	queue_work(vfdi_workqueue, &nic_data->peer_work);
1489
1490	return 0;
1491}
1492
1493void efx_siena_sriov_tx_flush_done(struct efx_nic *efx, efx_qword_t *event)
1494{
1495	struct siena_vf *vf;
1496	unsigned queue, qid;
1497
1498	queue = EFX_QWORD_FIELD(*event,  FSF_AZ_DRIVER_EV_SUBDATA);
1499	if (map_vi_index(efx, queue, &vf, &qid))
1500		return;
1501	/* Ignore flush completions triggered by an FLR */
1502	if (!test_bit(qid, vf->txq_mask))
1503		return;
1504
1505	__clear_bit(qid, vf->txq_mask);
1506	--vf->txq_count;
1507
1508	if (efx_vfdi_flush_wake(vf))
1509		wake_up(&vf->flush_waitq);
1510}
1511
1512void efx_siena_sriov_rx_flush_done(struct efx_nic *efx, efx_qword_t *event)
1513{
1514	struct siena_vf *vf;
1515	unsigned ev_failed, queue, qid;
1516
1517	queue = EFX_QWORD_FIELD(*event, FSF_AZ_DRIVER_EV_RX_DESCQ_ID);
1518	ev_failed = EFX_QWORD_FIELD(*event,
1519				    FSF_AZ_DRIVER_EV_RX_FLUSH_FAIL);
1520	if (map_vi_index(efx, queue, &vf, &qid))
1521		return;
1522	if (!test_bit(qid, vf->rxq_mask))
1523		return;
1524
1525	if (ev_failed) {
1526		set_bit(qid, vf->rxq_retry_mask);
1527		atomic_inc(&vf->rxq_retry_count);
1528	} else {
1529		__clear_bit(qid, vf->rxq_mask);
1530		--vf->rxq_count;
1531	}
1532	if (efx_vfdi_flush_wake(vf))
1533		wake_up(&vf->flush_waitq);
1534}
1535
1536/* Called from napi. Schedule the reset work item */
1537void efx_siena_sriov_desc_fetch_err(struct efx_nic *efx, unsigned dmaq)
1538{
1539	struct siena_vf *vf;
1540	unsigned int rel;
1541
1542	if (map_vi_index(efx, dmaq, &vf, &rel))
1543		return;
1544
1545	if (net_ratelimit())
1546		netif_err(efx, hw, efx->net_dev,
1547			  "VF %d DMA Q %d reports descriptor fetch error.\n",
1548			  vf->index, rel);
1549	queue_work(vfdi_workqueue, &vf->reset_work);
1550}
1551
1552/* Reset all VFs */
1553void efx_siena_sriov_reset(struct efx_nic *efx)
1554{
1555	struct siena_nic_data *nic_data = efx->nic_data;
1556	unsigned int vf_i;
1557	struct efx_buffer buf;
1558	struct siena_vf *vf;
1559
1560	ASSERT_RTNL();
1561
1562	if (efx->vf_init_count == 0)
1563		return;
1564
1565	efx_siena_sriov_usrev(efx, true);
1566	(void)efx_siena_sriov_cmd(efx, true, NULL, NULL);
1567
1568	if (efx_nic_alloc_buffer(efx, &buf, EFX_PAGE_SIZE, GFP_NOIO))
1569		return;
1570
1571	for (vf_i = 0; vf_i < efx->vf_init_count; ++vf_i) {
1572		vf = nic_data->vf + vf_i;
1573		efx_siena_sriov_reset_vf(vf, &buf);
1574	}
1575
1576	efx_nic_free_buffer(efx, &buf);
1577}
1578
1579int efx_init_sriov(void)
1580{
1581	/* A single threaded workqueue is sufficient. efx_siena_sriov_vfdi() and
1582	 * efx_siena_sriov_peer_work() spend almost all their time sleeping for
1583	 * MCDI to complete anyway
1584	 */
1585	vfdi_workqueue = create_singlethread_workqueue("sfc_vfdi");
1586	if (!vfdi_workqueue)
1587		return -ENOMEM;
1588	return 0;
1589}
1590
1591void efx_fini_sriov(void)
1592{
1593	destroy_workqueue(vfdi_workqueue);
1594}
1595
1596int efx_siena_sriov_set_vf_mac(struct efx_nic *efx, int vf_i, u8 *mac)
1597{
1598	struct siena_nic_data *nic_data = efx->nic_data;
1599	struct siena_vf *vf;
1600
1601	if (vf_i >= efx->vf_init_count)
1602		return -EINVAL;
1603	vf = nic_data->vf + vf_i;
1604
1605	mutex_lock(&vf->status_lock);
1606	ether_addr_copy(vf->addr.mac_addr, mac);
1607	__efx_siena_sriov_update_vf_addr(vf);
1608	mutex_unlock(&vf->status_lock);
1609
1610	return 0;
1611}
1612
1613int efx_siena_sriov_set_vf_vlan(struct efx_nic *efx, int vf_i,
1614				u16 vlan, u8 qos)
1615{
1616	struct siena_nic_data *nic_data = efx->nic_data;
1617	struct siena_vf *vf;
1618	u16 tci;
1619
1620	if (vf_i >= efx->vf_init_count)
1621		return -EINVAL;
1622	vf = nic_data->vf + vf_i;
1623
1624	mutex_lock(&vf->status_lock);
1625	tci = (vlan & VLAN_VID_MASK) | ((qos & 0x7) << VLAN_PRIO_SHIFT);
1626	vf->addr.tci = htons(tci);
1627	__efx_siena_sriov_update_vf_addr(vf);
1628	mutex_unlock(&vf->status_lock);
1629
1630	return 0;
1631}
1632
1633int efx_siena_sriov_set_vf_spoofchk(struct efx_nic *efx, int vf_i,
1634				    bool spoofchk)
1635{
1636	struct siena_nic_data *nic_data = efx->nic_data;
1637	struct siena_vf *vf;
1638	int rc;
1639
1640	if (vf_i >= efx->vf_init_count)
1641		return -EINVAL;
1642	vf = nic_data->vf + vf_i;
1643
1644	mutex_lock(&vf->txq_lock);
1645	if (vf->txq_count == 0) {
1646		vf->tx_filter_mode =
1647			spoofchk ? VF_TX_FILTER_ON : VF_TX_FILTER_OFF;
1648		rc = 0;
1649	} else {
1650		/* This cannot be changed while TX queues are running */
1651		rc = -EBUSY;
1652	}
1653	mutex_unlock(&vf->txq_lock);
1654	return rc;
1655}
1656
1657int efx_siena_sriov_get_vf_config(struct efx_nic *efx, int vf_i,
1658				  struct ifla_vf_info *ivi)
1659{
1660	struct siena_nic_data *nic_data = efx->nic_data;
1661	struct siena_vf *vf;
1662	u16 tci;
1663
1664	if (vf_i >= efx->vf_init_count)
1665		return -EINVAL;
1666	vf = nic_data->vf + vf_i;
1667
1668	ivi->vf = vf_i;
1669	ether_addr_copy(ivi->mac, vf->addr.mac_addr);
1670	ivi->max_tx_rate = 0;
1671	ivi->min_tx_rate = 0;
1672	tci = ntohs(vf->addr.tci);
1673	ivi->vlan = tci & VLAN_VID_MASK;
1674	ivi->qos = (tci >> VLAN_PRIO_SHIFT) & 0x7;
1675	ivi->spoofchk = vf->tx_filter_mode == VF_TX_FILTER_ON;
1676
1677	return 0;
1678}
1679
1680bool efx_siena_sriov_wanted(struct efx_nic *efx)
1681{
1682	return efx->vf_count != 0;
1683}
1684
1685int efx_siena_sriov_configure(struct efx_nic *efx, int num_vfs)
1686{
1687	return 0;
1688}
1689