brintos

brintos / linux-shallow public Read only

0
0
Text · 46.7 KiB · 8353c15 Raw
1688 lines · c
1// SPDX-License-Identifier: GPL-2.0-only2/****************************************************************************3 * Driver for Solarflare network controllers and boards4 * Copyright 2010-2012 Solarflare Communications Inc.5 */6#include <linux/pci.h>7#include <linux/module.h>8#include "net_driver.h"9#include "efx.h"10#include "efx_channels.h"11#include "nic.h"12#include "io.h"13#include "mcdi.h"14#include "filter.h"15#include "mcdi_pcol.h"16#include "farch_regs.h"17#include "siena_sriov.h"18#include "vfdi.h"19 20/* Number of longs required to track all the VIs in a VF */21#define VI_MASK_LENGTH BITS_TO_LONGS(1 << EFX_VI_SCALE_MAX)22 23/* Maximum number of RX queues supported */24#define VF_MAX_RX_QUEUES 6325 26/**27 * enum efx_vf_tx_filter_mode - TX MAC filtering behaviour28 * @VF_TX_FILTER_OFF: Disabled29 * @VF_TX_FILTER_AUTO: Enabled if MAC address assigned to VF and only30 *	2 TX queues allowed per VF.31 * @VF_TX_FILTER_ON: Enabled32 */33enum efx_vf_tx_filter_mode {34	VF_TX_FILTER_OFF,35	VF_TX_FILTER_AUTO,36	VF_TX_FILTER_ON,37};38 39/**40 * struct siena_vf - Back-end resource and protocol state for a PCI VF41 * @efx: The Efx NIC owning this VF42 * @pci_rid: The PCI requester ID for this VF43 * @pci_name: The PCI name (formatted address) of this VF44 * @index: Index of VF within its port and PF.45 * @req: VFDI incoming request work item. Incoming USR_EV events are received46 *	by the NAPI handler, but must be handled by executing MCDI requests47 *	inside a work item.48 * @req_addr: VFDI incoming request DMA address (in VF's PCI address space).49 * @req_type: Expected next incoming (from VF) %VFDI_EV_TYPE member.50 * @req_seqno: Expected next incoming (from VF) %VFDI_EV_SEQ member.51 * @msg_seqno: Next %VFDI_EV_SEQ member to reply to VF. Protected by52 *	@status_lock53 * @busy: VFDI request queued to be processed or being processed. Receiving54 *	a VFDI request when @busy is set is an error condition.55 * @buf: Incoming VFDI requests are DMA from the VF into this buffer.56 * @buftbl_base: Buffer table entries for this VF start at this index.57 * @rx_filtering: Receive filtering has been requested by the VF driver.58 * @rx_filter_flags: The flags sent in the %VFDI_OP_INSERT_FILTER request.59 * @rx_filter_qid: VF relative qid for RX filter requested by VF.60 * @rx_filter_id: Receive MAC filter ID. Only one filter per VF is supported.61 * @tx_filter_mode: Transmit MAC filtering mode.62 * @tx_filter_id: Transmit MAC filter ID.63 * @addr: The MAC address and outer vlan tag of the VF.64 * @status_addr: VF DMA address of page for &struct vfdi_status updates.65 * @status_lock: Mutex protecting @msg_seqno, @status_addr, @addr,66 *	@peer_page_addrs and @peer_page_count from simultaneous67 *	updates by the VM and consumption by68 *	efx_siena_sriov_update_vf_addr()69 * @peer_page_addrs: Pointer to an array of guest pages for local addresses.70 * @peer_page_count: Number of entries in @peer_page_count.71 * @evq0_addrs: Array of guest pages backing evq0.72 * @evq0_count: Number of entries in @evq0_addrs.73 * @flush_waitq: wait queue used by %VFDI_OP_FINI_ALL_QUEUES handler74 *	to wait for flush completions.75 * @txq_lock: Mutex for TX queue allocation.76 * @txq_mask: Mask of initialized transmit queues.77 * @txq_count: Number of initialized transmit queues.78 * @rxq_mask: Mask of initialized receive queues.79 * @rxq_count: Number of initialized receive queues.80 * @rxq_retry_mask: Mask or receive queues that need to be flushed again81 *	due to flush failure.82 * @rxq_retry_count: Number of receive queues in @rxq_retry_mask.83 * @reset_work: Work item to schedule a VF reset.84 */85struct siena_vf {86	struct efx_nic *efx;87	unsigned int pci_rid;88	char pci_name[13]; /* dddd:bb:dd.f */89	unsigned int index;90	struct work_struct req;91	u64 req_addr;92	int req_type;93	unsigned req_seqno;94	unsigned msg_seqno;95	bool busy;96	struct efx_buffer buf;97	unsigned buftbl_base;98	bool rx_filtering;99	enum efx_filter_flags rx_filter_flags;100	unsigned rx_filter_qid;101	int rx_filter_id;102	enum efx_vf_tx_filter_mode tx_filter_mode;103	int tx_filter_id;104	struct vfdi_endpoint addr;105	u64 status_addr;106	struct mutex status_lock;107	u64 *peer_page_addrs;108	unsigned peer_page_count;109	u64 evq0_addrs[EFX_MAX_VF_EVQ_SIZE * sizeof(efx_qword_t) /110		       EFX_BUF_SIZE];111	unsigned evq0_count;112	wait_queue_head_t flush_waitq;113	struct mutex txq_lock;114	unsigned long txq_mask[VI_MASK_LENGTH];115	unsigned txq_count;116	unsigned long rxq_mask[VI_MASK_LENGTH];117	unsigned rxq_count;118	unsigned long rxq_retry_mask[VI_MASK_LENGTH];119	atomic_t rxq_retry_count;120	struct work_struct reset_work;121};122 123struct efx_memcpy_req {124	unsigned int from_rid;125	void *from_buf;126	u64 from_addr;127	unsigned int to_rid;128	u64 to_addr;129	unsigned length;130};131 132/**133 * struct efx_local_addr - A MAC address on the vswitch without a VF.134 *135 * Siena does not have a switch, so VFs can't transmit data to each136 * other. Instead the VFs must be made aware of the local addresses137 * on the vswitch, so that they can arrange for an alternative138 * software datapath to be used.139 *140 * @link: List head for insertion into efx->local_addr_list.141 * @addr: Ethernet address142 */143struct efx_local_addr {144	struct list_head link;145	u8 addr[ETH_ALEN];146};147 148/**149 * struct efx_endpoint_page - Page of vfdi_endpoint structures150 *151 * @link: List head for insertion into efx->local_page_list.152 * @ptr: Pointer to page.153 * @addr: DMA address of page.154 */155struct efx_endpoint_page {156	struct list_head link;157	void *ptr;158	dma_addr_t addr;159};160 161/* Buffer table entries are reserved txq0,rxq0,evq0,txq1,rxq1,evq1 */162#define EFX_BUFTBL_TXQ_BASE(_vf, _qid)					\163	((_vf)->buftbl_base + EFX_VF_BUFTBL_PER_VI * (_qid))164#define EFX_BUFTBL_RXQ_BASE(_vf, _qid)					\165	(EFX_BUFTBL_TXQ_BASE(_vf, _qid) +				\166	 (EFX_MAX_DMAQ_SIZE * sizeof(efx_qword_t) / EFX_BUF_SIZE))167#define EFX_BUFTBL_EVQ_BASE(_vf, _qid)					\168	(EFX_BUFTBL_TXQ_BASE(_vf, _qid) +				\169	 (2 * EFX_MAX_DMAQ_SIZE * sizeof(efx_qword_t) / EFX_BUF_SIZE))170 171#define EFX_FIELD_MASK(_field)			\172	((1 << _field ## _WIDTH) - 1)173 174/* VFs can only use this many transmit channels */175static unsigned int vf_max_tx_channels = 2;176module_param(vf_max_tx_channels, uint, 0444);177MODULE_PARM_DESC(vf_max_tx_channels,178		 "Limit the number of TX channels VFs can use");179 180static int max_vfs = -1;181module_param(max_vfs, int, 0444);182MODULE_PARM_DESC(max_vfs,183		 "Reduce the number of VFs initialized by the driver");184 185/* Workqueue used by VFDI communication.  We can't use the global186 * workqueue because it may be running the VF driver's probe()187 * routine, which will be blocked there waiting for a VFDI response.188 */189static struct workqueue_struct *vfdi_workqueue;190 191static unsigned abs_index(struct siena_vf *vf, unsigned index)192{193	return EFX_VI_BASE + vf->index * efx_vf_size(vf->efx) + index;194}195 196static int efx_siena_sriov_cmd(struct efx_nic *efx, bool enable,197			       unsigned *vi_scale_out, unsigned *vf_total_out)198{199	MCDI_DECLARE_BUF(inbuf, MC_CMD_SRIOV_IN_LEN);200	MCDI_DECLARE_BUF(outbuf, MC_CMD_SRIOV_OUT_LEN);201	unsigned vi_scale, vf_total;202	size_t outlen;203	int rc;204 205	MCDI_SET_DWORD(inbuf, SRIOV_IN_ENABLE, enable ? 1 : 0);206	MCDI_SET_DWORD(inbuf, SRIOV_IN_VI_BASE, EFX_VI_BASE);207	MCDI_SET_DWORD(inbuf, SRIOV_IN_VF_COUNT, efx->vf_count);208 209	rc = efx_siena_mcdi_rpc_quiet(efx, MC_CMD_SRIOV, inbuf,210				      MC_CMD_SRIOV_IN_LEN, outbuf,211				      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_siena_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 modified300 * underneath the feet of the VF301 */302static void efx_siena_sriov_reset_tx_filter(struct siena_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 explicitly321	 * 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 siena_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 siena_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 interlock396 * with VFDI requests, and they must be serialised against manipulation of397 * local_page_list, either by acquiring local_lock or by running from398 * efx_siena_sriov_peer_work()399 */400static void __efx_siena_sriov_push_vf_status(struct siena_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). This425	 * assumes that the non-generation portion of vfdi_status is in426	 * 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 more441			 * 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 siena_vf **vf_out, unsigned *rel_index_out)513{514	struct siena_nic_data *nic_data = efx->nic_data;515	unsigned vf_i;516 517	if (abs_index < EFX_VI_BASE)518		return true;519	vf_i = (abs_index - EFX_VI_BASE) / efx_vf_size(efx);520	if (vf_i >= efx->vf_init_count)521		return true;522 523	if (vf_out)524		*vf_out = nic_data->vf + vf_i;525	if (rel_index_out)526		*rel_index_out = abs_index % efx_vf_size(efx);527	return false;528}529 530static int efx_vfdi_init_evq(struct siena_vf *vf)531{532	struct efx_nic *efx = vf->efx;533	struct vfdi_req *req = vf->buf.addr;534	unsigned vf_evq = req->u.init_evq.index;535	unsigned buf_count = req->u.init_evq.buf_count;536	unsigned abs_evq = abs_index(vf, vf_evq);537	unsigned buftbl = EFX_BUFTBL_EVQ_BASE(vf, vf_evq);538	efx_oword_t reg;539 540	if (bad_vf_index(efx, vf_evq) ||541	    bad_buf_count(buf_count, EFX_MAX_VF_EVQ_SIZE)) {542		if (net_ratelimit())543			netif_err(efx, hw, efx->net_dev,544				  "ERROR: Invalid INIT_EVQ from %s: evq %d bufs %d\n",545				  vf->pci_name, vf_evq, buf_count);546		return VFDI_RC_EINVAL;547	}548 549	efx_siena_sriov_bufs(efx, buftbl, req->u.init_evq.addr, buf_count);550 551	EFX_POPULATE_OWORD_3(reg,552			     FRF_CZ_TIMER_Q_EN, 1,553			     FRF_CZ_HOST_NOTIFY_MODE, 0,554			     FRF_CZ_TIMER_MODE, FFE_CZ_TIMER_MODE_DIS);555	efx_writeo_table(efx, &reg, FR_BZ_TIMER_TBL, abs_evq);556	EFX_POPULATE_OWORD_3(reg,557			     FRF_AZ_EVQ_EN, 1,558			     FRF_AZ_EVQ_SIZE, __ffs(buf_count),559			     FRF_AZ_EVQ_BUF_BASE_ID, buftbl);560	efx_writeo_table(efx, &reg, FR_BZ_EVQ_PTR_TBL, abs_evq);561 562	if (vf_evq == 0) {563		memcpy(vf->evq0_addrs, req->u.init_evq.addr,564		       buf_count * sizeof(u64));565		vf->evq0_count = buf_count;566	}567 568	return VFDI_RC_SUCCESS;569}570 571static int efx_vfdi_init_rxq(struct siena_vf *vf)572{573	struct efx_nic *efx = vf->efx;574	struct vfdi_req *req = vf->buf.addr;575	unsigned vf_rxq = req->u.init_rxq.index;576	unsigned vf_evq = req->u.init_rxq.evq;577	unsigned buf_count = req->u.init_rxq.buf_count;578	unsigned buftbl = EFX_BUFTBL_RXQ_BASE(vf, vf_rxq);579	unsigned label;580	efx_oword_t reg;581 582	if (bad_vf_index(efx, vf_evq) || bad_vf_index(efx, vf_rxq) ||583	    vf_rxq >= VF_MAX_RX_QUEUES ||584	    bad_buf_count(buf_count, EFX_MAX_DMAQ_SIZE)) {585		if (net_ratelimit())586			netif_err(efx, hw, efx->net_dev,587				  "ERROR: Invalid INIT_RXQ from %s: rxq %d evq %d "588				  "buf_count %d\n", vf->pci_name, vf_rxq,589				  vf_evq, buf_count);590		return VFDI_RC_EINVAL;591	}592	if (__test_and_set_bit(req->u.init_rxq.index, vf->rxq_mask))593		++vf->rxq_count;594	efx_siena_sriov_bufs(efx, buftbl, req->u.init_rxq.addr, buf_count);595 596	label = req->u.init_rxq.label & EFX_FIELD_MASK(FRF_AZ_RX_DESCQ_LABEL);597	EFX_POPULATE_OWORD_6(reg,598			     FRF_AZ_RX_DESCQ_BUF_BASE_ID, buftbl,599			     FRF_AZ_RX_DESCQ_EVQ_ID, abs_index(vf, vf_evq),600			     FRF_AZ_RX_DESCQ_LABEL, label,601			     FRF_AZ_RX_DESCQ_SIZE, __ffs(buf_count),602			     FRF_AZ_RX_DESCQ_JUMBO,603			     !!(req->u.init_rxq.flags &604				VFDI_RXQ_FLAG_SCATTER_EN),605			     FRF_AZ_RX_DESCQ_EN, 1);606	efx_writeo_table(efx, &reg, FR_BZ_RX_DESC_PTR_TBL,607			 abs_index(vf, vf_rxq));608 609	return VFDI_RC_SUCCESS;610}611 612static int efx_vfdi_init_txq(struct siena_vf *vf)613{614	struct efx_nic *efx = vf->efx;615	struct vfdi_req *req = vf->buf.addr;616	unsigned vf_txq = req->u.init_txq.index;617	unsigned vf_evq = req->u.init_txq.evq;618	unsigned buf_count = req->u.init_txq.buf_count;619	unsigned buftbl = EFX_BUFTBL_TXQ_BASE(vf, vf_txq);620	unsigned label, eth_filt_en;621	efx_oword_t reg;622 623	if (bad_vf_index(efx, vf_evq) || bad_vf_index(efx, vf_txq) ||624	    vf_txq >= vf_max_tx_channels ||625	    bad_buf_count(buf_count, EFX_MAX_DMAQ_SIZE)) {626		if (net_ratelimit())627			netif_err(efx, hw, efx->net_dev,628				  "ERROR: Invalid INIT_TXQ from %s: txq %d evq %d "629				  "buf_count %d\n", vf->pci_name, vf_txq,630				  vf_evq, buf_count);631		return VFDI_RC_EINVAL;632	}633 634	mutex_lock(&vf->txq_lock);635	if (__test_and_set_bit(req->u.init_txq.index, vf->txq_mask))636		++vf->txq_count;637	mutex_unlock(&vf->txq_lock);638	efx_siena_sriov_bufs(efx, buftbl, req->u.init_txq.addr, buf_count);639 640	eth_filt_en = vf->tx_filter_mode == VF_TX_FILTER_ON;641 642	label = req->u.init_txq.label & EFX_FIELD_MASK(FRF_AZ_TX_DESCQ_LABEL);643	EFX_POPULATE_OWORD_8(reg,644			     FRF_CZ_TX_DPT_Q_MASK_WIDTH, min(efx->vi_scale, 1U),645			     FRF_CZ_TX_DPT_ETH_FILT_EN, eth_filt_en,646			     FRF_AZ_TX_DESCQ_EN, 1,647			     FRF_AZ_TX_DESCQ_BUF_BASE_ID, buftbl,648			     FRF_AZ_TX_DESCQ_EVQ_ID, abs_index(vf, vf_evq),649			     FRF_AZ_TX_DESCQ_LABEL, label,650			     FRF_AZ_TX_DESCQ_SIZE, __ffs(buf_count),651			     FRF_BZ_TX_NON_IP_DROP_DIS, 1);652	efx_writeo_table(efx, &reg, FR_BZ_TX_DESC_PTR_TBL,653			 abs_index(vf, vf_txq));654 655	return VFDI_RC_SUCCESS;656}657 658/* Returns true when efx_vfdi_fini_all_queues should wake */659static bool efx_vfdi_flush_wake(struct siena_vf *vf)660{661	/* Ensure that all updates are visible to efx_vfdi_fini_all_queues() */662	smp_mb();663 664	return (!vf->txq_count && !vf->rxq_count) ||665		atomic_read(&vf->rxq_retry_count);666}667 668static void efx_vfdi_flush_clear(struct siena_vf *vf)669{670	memset(vf->txq_mask, 0, sizeof(vf->txq_mask));671	vf->txq_count = 0;672	memset(vf->rxq_mask, 0, sizeof(vf->rxq_mask));673	vf->rxq_count = 0;674	memset(vf->rxq_retry_mask, 0, sizeof(vf->rxq_retry_mask));675	atomic_set(&vf->rxq_retry_count, 0);676}677 678static int efx_vfdi_fini_all_queues(struct siena_vf *vf)679{680	struct efx_nic *efx = vf->efx;681	efx_oword_t reg;682	unsigned count = efx_vf_size(efx);683	unsigned vf_offset = EFX_VI_BASE + vf->index * efx_vf_size(efx);684	unsigned timeout = HZ;685	unsigned index, rxqs_count;686	MCDI_DECLARE_BUF(inbuf, MC_CMD_FLUSH_RX_QUEUES_IN_LENMAX);687	int rc;688 689	BUILD_BUG_ON(VF_MAX_RX_QUEUES >690		     MC_CMD_FLUSH_RX_QUEUES_IN_QID_OFST_MAXNUM);691 692	rtnl_lock();693	efx_siena_prepare_flush(efx);694	rtnl_unlock();695 696	/* Flush all the initialized queues */697	rxqs_count = 0;698	for (index = 0; index < count; ++index) {699		if (test_bit(index, vf->txq_mask)) {700			EFX_POPULATE_OWORD_2(reg,701					     FRF_AZ_TX_FLUSH_DESCQ_CMD, 1,702					     FRF_AZ_TX_FLUSH_DESCQ,703					     vf_offset + index);704			efx_writeo(efx, &reg, FR_AZ_TX_FLUSH_DESCQ);705		}706		if (test_bit(index, vf->rxq_mask)) {707			MCDI_SET_ARRAY_DWORD(708				inbuf, FLUSH_RX_QUEUES_IN_QID_OFST,709				rxqs_count, vf_offset + index);710			rxqs_count++;711		}712	}713 714	atomic_set(&vf->rxq_retry_count, 0);715	while (timeout && (vf->rxq_count || vf->txq_count)) {716		rc = efx_siena_mcdi_rpc(efx, MC_CMD_FLUSH_RX_QUEUES, inbuf,717				  MC_CMD_FLUSH_RX_QUEUES_IN_LEN(rxqs_count),718				  NULL, 0, NULL);719		WARN_ON(rc < 0);720 721		timeout = wait_event_timeout(vf->flush_waitq,722					     efx_vfdi_flush_wake(vf),723					     timeout);724		rxqs_count = 0;725		for (index = 0; index < count; ++index) {726			if (test_and_clear_bit(index, vf->rxq_retry_mask)) {727				atomic_dec(&vf->rxq_retry_count);728				MCDI_SET_ARRAY_DWORD(729					inbuf, FLUSH_RX_QUEUES_IN_QID_OFST,730					rxqs_count, vf_offset + index);731				rxqs_count++;732			}733		}734	}735 736	rtnl_lock();737	siena_finish_flush(efx);738	rtnl_unlock();739 740	/* Irrespective of success/failure, fini the queues */741	EFX_ZERO_OWORD(reg);742	for (index = 0; index < count; ++index) {743		efx_writeo_table(efx, &reg, FR_BZ_RX_DESC_PTR_TBL,744				 vf_offset + index);745		efx_writeo_table(efx, &reg, FR_BZ_TX_DESC_PTR_TBL,746				 vf_offset + index);747		efx_writeo_table(efx, &reg, FR_BZ_EVQ_PTR_TBL,748				 vf_offset + index);749		efx_writeo_table(efx, &reg, FR_BZ_TIMER_TBL,750				 vf_offset + index);751	}752	efx_siena_sriov_bufs(efx, vf->buftbl_base, NULL,753			     EFX_VF_BUFTBL_PER_VI * efx_vf_size(efx));754	efx_vfdi_flush_clear(vf);755 756	vf->evq0_count = 0;757 758	return timeout ? 0 : VFDI_RC_ETIMEDOUT;759}760 761static int efx_vfdi_insert_filter(struct siena_vf *vf)762{763	struct efx_nic *efx = vf->efx;764	struct siena_nic_data *nic_data = efx->nic_data;765	struct vfdi_req *req = vf->buf.addr;766	unsigned vf_rxq = req->u.mac_filter.rxq;767	unsigned flags;768 769	if (bad_vf_index(efx, vf_rxq) || vf->rx_filtering) {770		if (net_ratelimit())771			netif_err(efx, hw, efx->net_dev,772				  "ERROR: Invalid INSERT_FILTER from %s: rxq %d "773				  "flags 0x%x\n", vf->pci_name, vf_rxq,774				  req->u.mac_filter.flags);775		return VFDI_RC_EINVAL;776	}777 778	flags = 0;779	if (req->u.mac_filter.flags & VFDI_MAC_FILTER_FLAG_RSS)780		flags |= EFX_FILTER_FLAG_RX_RSS;781	if (req->u.mac_filter.flags & VFDI_MAC_FILTER_FLAG_SCATTER)782		flags |= EFX_FILTER_FLAG_RX_SCATTER;783	vf->rx_filter_flags = flags;784	vf->rx_filter_qid = vf_rxq;785	vf->rx_filtering = true;786 787	efx_siena_sriov_reset_rx_filter(vf);788	queue_work(vfdi_workqueue, &nic_data->peer_work);789 790	return VFDI_RC_SUCCESS;791}792 793static int efx_vfdi_remove_all_filters(struct siena_vf *vf)794{795	struct efx_nic *efx = vf->efx;796	struct siena_nic_data *nic_data = efx->nic_data;797 798	vf->rx_filtering = false;799	efx_siena_sriov_reset_rx_filter(vf);800	queue_work(vfdi_workqueue, &nic_data->peer_work);801 802	return VFDI_RC_SUCCESS;803}804 805static int efx_vfdi_set_status_page(struct siena_vf *vf)806{807	struct efx_nic *efx = vf->efx;808	struct siena_nic_data *nic_data = efx->nic_data;809	struct vfdi_req *req = vf->buf.addr;810	u64 page_count = req->u.set_status_page.peer_page_count;811	u64 max_page_count =812		(EFX_PAGE_SIZE -813		 offsetof(struct vfdi_req, u.set_status_page.peer_page_addr[0]))814		/ sizeof(req->u.set_status_page.peer_page_addr[0]);815 816	if (!req->u.set_status_page.dma_addr || page_count > max_page_count) {817		if (net_ratelimit())818			netif_err(efx, hw, efx->net_dev,819				  "ERROR: Invalid SET_STATUS_PAGE from %s\n",820				  vf->pci_name);821		return VFDI_RC_EINVAL;822	}823 824	mutex_lock(&nic_data->local_lock);825	mutex_lock(&vf->status_lock);826	vf->status_addr = req->u.set_status_page.dma_addr;827 828	kfree(vf->peer_page_addrs);829	vf->peer_page_addrs = NULL;830	vf->peer_page_count = 0;831 832	if (page_count) {833		vf->peer_page_addrs = kcalloc(page_count, sizeof(u64),834					      GFP_KERNEL);835		if (vf->peer_page_addrs) {836			memcpy(vf->peer_page_addrs,837			       req->u.set_status_page.peer_page_addr,838			       page_count * sizeof(u64));839			vf->peer_page_count = page_count;840		}841	}842 843	__efx_siena_sriov_push_vf_status(vf);844	mutex_unlock(&vf->status_lock);845	mutex_unlock(&nic_data->local_lock);846 847	return VFDI_RC_SUCCESS;848}849 850static int efx_vfdi_clear_status_page(struct siena_vf *vf)851{852	mutex_lock(&vf->status_lock);853	vf->status_addr = 0;854	mutex_unlock(&vf->status_lock);855 856	return VFDI_RC_SUCCESS;857}858 859typedef int (*efx_vfdi_op_t)(struct siena_vf *vf);860 861static const efx_vfdi_op_t vfdi_ops[VFDI_OP_LIMIT] = {862	[VFDI_OP_INIT_EVQ] = efx_vfdi_init_evq,863	[VFDI_OP_INIT_TXQ] = efx_vfdi_init_txq,864	[VFDI_OP_INIT_RXQ] = efx_vfdi_init_rxq,865	[VFDI_OP_FINI_ALL_QUEUES] = efx_vfdi_fini_all_queues,866	[VFDI_OP_INSERT_FILTER] = efx_vfdi_insert_filter,867	[VFDI_OP_REMOVE_ALL_FILTERS] = efx_vfdi_remove_all_filters,868	[VFDI_OP_SET_STATUS_PAGE] = efx_vfdi_set_status_page,869	[VFDI_OP_CLEAR_STATUS_PAGE] = efx_vfdi_clear_status_page,870};871 872static void efx_siena_sriov_vfdi(struct work_struct *work)873{874	struct siena_vf *vf = container_of(work, struct siena_vf, req);875	struct efx_nic *efx = vf->efx;876	struct vfdi_req *req = vf->buf.addr;877	struct efx_memcpy_req copy[2];878	int rc;879 880	/* Copy this page into the local address space */881	memset(copy, '\0', sizeof(copy));882	copy[0].from_rid = vf->pci_rid;883	copy[0].from_addr = vf->req_addr;884	copy[0].to_rid = efx->pci_dev->devfn;885	copy[0].to_addr = vf->buf.dma_addr;886	copy[0].length = EFX_PAGE_SIZE;887	rc = efx_siena_sriov_memcpy(efx, copy, 1);888	if (rc) {889		/* If we can't get the request, we can't reply to the caller */890		if (net_ratelimit())891			netif_err(efx, hw, efx->net_dev,892				  "ERROR: Unable to fetch VFDI request from %s rc %d\n",893				  vf->pci_name, -rc);894		vf->busy = false;895		return;896	}897 898	if (req->op < VFDI_OP_LIMIT && vfdi_ops[req->op] != NULL) {899		rc = vfdi_ops[req->op](vf);900		if (rc == 0) {901			netif_dbg(efx, hw, efx->net_dev,902				  "vfdi request %d from %s ok\n",903				  req->op, vf->pci_name);904		}905	} else {906		netif_dbg(efx, hw, efx->net_dev,907			  "ERROR: Unrecognised request %d from VF %s addr "908			  "%llx\n", req->op, vf->pci_name,909			  (unsigned long long)vf->req_addr);910		rc = VFDI_RC_EOPNOTSUPP;911	}912 913	/* Allow subsequent VF requests */914	vf->busy = false;915	smp_wmb();916 917	/* Respond to the request */918	req->rc = rc;919	req->op = VFDI_OP_RESPONSE;920 921	memset(copy, '\0', sizeof(copy));922	copy[0].from_buf = &req->rc;923	copy[0].to_rid = vf->pci_rid;924	copy[0].to_addr = vf->req_addr + offsetof(struct vfdi_req, rc);925	copy[0].length = sizeof(req->rc);926	copy[1].from_buf = &req->op;927	copy[1].to_rid = vf->pci_rid;928	copy[1].to_addr = vf->req_addr + offsetof(struct vfdi_req, op);929	copy[1].length = sizeof(req->op);930 931	(void)efx_siena_sriov_memcpy(efx, copy, ARRAY_SIZE(copy));932}933 934 935 936/* After a reset the event queues inside the guests no longer exist. Fill the937 * event ring in guest memory with VFDI reset events, then (re-initialise) the938 * event queue to raise an interrupt. The guest driver will then recover.939 */940 941static void efx_siena_sriov_reset_vf(struct siena_vf *vf,942				     struct efx_buffer *buffer)943{944	struct efx_nic *efx = vf->efx;945	struct efx_memcpy_req copy_req[4];946	efx_qword_t event;947	unsigned int pos, count, k, buftbl, abs_evq;948	efx_oword_t reg;949	efx_dword_t ptr;950	int rc;951 952	BUG_ON(buffer->len != EFX_PAGE_SIZE);953 954	if (!vf->evq0_count)955		return;956	BUG_ON(vf->evq0_count & (vf->evq0_count - 1));957 958	mutex_lock(&vf->status_lock);959	EFX_POPULATE_QWORD_3(event,960			     FSF_AZ_EV_CODE, FSE_CZ_EV_CODE_USER_EV,961			     VFDI_EV_SEQ, vf->msg_seqno,962			     VFDI_EV_TYPE, VFDI_EV_TYPE_RESET);963	vf->msg_seqno++;964	for (pos = 0; pos < EFX_PAGE_SIZE; pos += sizeof(event))965		memcpy(buffer->addr + pos, &event, sizeof(event));966 967	for (pos = 0; pos < vf->evq0_count; pos += count) {968		count = min_t(unsigned, vf->evq0_count - pos,969			      ARRAY_SIZE(copy_req));970		for (k = 0; k < count; k++) {971			copy_req[k].from_buf = NULL;972			copy_req[k].from_rid = efx->pci_dev->devfn;973			copy_req[k].from_addr = buffer->dma_addr;974			copy_req[k].to_rid = vf->pci_rid;975			copy_req[k].to_addr = vf->evq0_addrs[pos + k];976			copy_req[k].length = EFX_PAGE_SIZE;977		}978		rc = efx_siena_sriov_memcpy(efx, copy_req, count);979		if (rc) {980			if (net_ratelimit())981				netif_err(efx, hw, efx->net_dev,982					  "ERROR: Unable to notify %s of reset"983					  ": %d\n", vf->pci_name, -rc);984			break;985		}986	}987 988	/* Reinitialise, arm and trigger evq0 */989	abs_evq = abs_index(vf, 0);990	buftbl = EFX_BUFTBL_EVQ_BASE(vf, 0);991	efx_siena_sriov_bufs(efx, buftbl, vf->evq0_addrs, vf->evq0_count);992 993	EFX_POPULATE_OWORD_3(reg,994			     FRF_CZ_TIMER_Q_EN, 1,995			     FRF_CZ_HOST_NOTIFY_MODE, 0,996			     FRF_CZ_TIMER_MODE, FFE_CZ_TIMER_MODE_DIS);997	efx_writeo_table(efx, &reg, FR_BZ_TIMER_TBL, abs_evq);998	EFX_POPULATE_OWORD_3(reg,999			     FRF_AZ_EVQ_EN, 1,1000			     FRF_AZ_EVQ_SIZE, __ffs(vf->evq0_count),1001			     FRF_AZ_EVQ_BUF_BASE_ID, buftbl);1002	efx_writeo_table(efx, &reg, FR_BZ_EVQ_PTR_TBL, abs_evq);1003	EFX_POPULATE_DWORD_1(ptr, FRF_AZ_EVQ_RPTR, 0);1004	efx_writed(efx, &ptr, FR_BZ_EVQ_RPTR + FR_BZ_EVQ_RPTR_STEP * abs_evq);1005 1006	mutex_unlock(&vf->status_lock);1007}1008 1009static void efx_siena_sriov_reset_vf_work(struct work_struct *work)1010{1011	struct siena_vf *vf = container_of(work, struct siena_vf, req);1012	struct efx_nic *efx = vf->efx;1013	struct efx_buffer buf;1014 1015	if (!efx_siena_alloc_buffer(efx, &buf, EFX_PAGE_SIZE, GFP_NOIO)) {1016		efx_siena_sriov_reset_vf(vf, &buf);1017		efx_siena_free_buffer(efx, &buf);1018	}1019}1020 1021static void efx_siena_sriov_handle_no_channel(struct efx_nic *efx)1022{1023	netif_err(efx, drv, efx->net_dev,1024		  "ERROR: IOV requires MSI-X and 1 additional interrupt"1025		  "vector. IOV disabled\n");1026	efx->vf_count = 0;1027}1028 1029static int efx_siena_sriov_probe_channel(struct efx_channel *channel)1030{1031	struct siena_nic_data *nic_data = channel->efx->nic_data;1032	nic_data->vfdi_channel = channel;1033 1034	return 0;1035}1036 1037static void1038efx_siena_sriov_get_channel_name(struct efx_channel *channel,1039				 char *buf, size_t len)1040{1041	snprintf(buf, len, "%s-iov", channel->efx->name);1042}1043 1044static const struct efx_channel_type efx_siena_sriov_channel_type = {1045	.handle_no_channel	= efx_siena_sriov_handle_no_channel,1046	.pre_probe		= efx_siena_sriov_probe_channel,1047	.post_remove		= efx_siena_channel_dummy_op_void,1048	.get_name		= efx_siena_sriov_get_channel_name,1049	/* no copy operation; channel must not be reallocated */1050	.keep_eventq		= true,1051};1052 1053void efx_siena_sriov_probe(struct efx_nic *efx)1054{1055	unsigned count;1056 1057	if (!max_vfs)1058		return;1059 1060	if (efx_siena_sriov_cmd(efx, false, &efx->vi_scale, &count)) {1061		pci_info(efx->pci_dev, "no SR-IOV VFs probed\n");1062		return;1063	}1064	if (count > 0 && count > max_vfs)1065		count = max_vfs;1066 1067	/* efx_nic_dimension_resources() will reduce vf_count as appopriate */1068	efx->vf_count = count;1069 1070	efx->extra_channel_type[EFX_EXTRA_CHANNEL_IOV] = &efx_siena_sriov_channel_type;1071}1072 1073/* Copy the list of individual addresses into the vfdi_status.peers1074 * array and auxiliary pages, protected by %local_lock. Drop that lock1075 * and then broadcast the address list to every VF.1076 */1077static void efx_siena_sriov_peer_work(struct work_struct *data)1078{1079	struct siena_nic_data *nic_data = container_of(data,1080						       struct siena_nic_data,1081						       peer_work);1082	struct efx_nic *efx = nic_data->efx;1083	struct vfdi_status *vfdi_status = nic_data->vfdi_status.addr;1084	struct siena_vf *vf;1085	struct efx_local_addr *local_addr;1086	struct vfdi_endpoint *peer;1087	struct efx_endpoint_page *epp;1088	struct list_head pages;1089	unsigned int peer_space;1090	unsigned int peer_count;1091	unsigned int pos;1092 1093	mutex_lock(&nic_data->local_lock);1094 1095	/* Move the existing peer pages off %local_page_list */1096	INIT_LIST_HEAD(&pages);1097	list_splice_tail_init(&nic_data->local_page_list, &pages);1098 1099	/* Populate the VF addresses starting from entry 1 (entry 0 is1100	 * the PF address)1101	 */1102	peer = vfdi_status->peers + 1;1103	peer_space = ARRAY_SIZE(vfdi_status->peers) - 1;1104	peer_count = 1;1105	for (pos = 0; pos < efx->vf_count; ++pos) {1106		vf = nic_data->vf + pos;1107 1108		mutex_lock(&vf->status_lock);1109		if (vf->rx_filtering && !is_zero_ether_addr(vf->addr.mac_addr)) {1110			*peer++ = vf->addr;1111			++peer_count;1112			--peer_space;1113			BUG_ON(peer_space == 0);1114		}1115		mutex_unlock(&vf->status_lock);1116	}1117 1118	/* Fill the remaining addresses */1119	list_for_each_entry(local_addr, &nic_data->local_addr_list, link) {1120		ether_addr_copy(peer->mac_addr, local_addr->addr);1121		peer->tci = 0;1122		++peer;1123		++peer_count;1124		if (--peer_space == 0) {1125			if (list_empty(&pages)) {1126				epp = kmalloc(sizeof(*epp), GFP_KERNEL);1127				if (!epp)1128					break;1129				epp->ptr = dma_alloc_coherent(1130					&efx->pci_dev->dev, EFX_PAGE_SIZE,1131					&epp->addr, GFP_KERNEL);1132				if (!epp->ptr) {1133					kfree(epp);1134					break;1135				}1136			} else {1137				epp = list_first_entry(1138					&pages, struct efx_endpoint_page, link);1139				list_del(&epp->link);1140			}1141 1142			list_add_tail(&epp->link, &nic_data->local_page_list);1143			peer = (struct vfdi_endpoint *)epp->ptr;1144			peer_space = EFX_PAGE_SIZE / sizeof(struct vfdi_endpoint);1145		}1146	}1147	vfdi_status->peer_count = peer_count;1148	mutex_unlock(&nic_data->local_lock);1149 1150	/* Free any now unused endpoint pages */1151	while (!list_empty(&pages)) {1152		epp = list_first_entry(1153			&pages, struct efx_endpoint_page, link);1154		list_del(&epp->link);1155		dma_free_coherent(&efx->pci_dev->dev, EFX_PAGE_SIZE,1156				  epp->ptr, epp->addr);1157		kfree(epp);1158	}1159 1160	/* Finally, push the pages */1161	for (pos = 0; pos < efx->vf_count; ++pos) {1162		vf = nic_data->vf + pos;1163 1164		mutex_lock(&vf->status_lock);1165		if (vf->status_addr)1166			__efx_siena_sriov_push_vf_status(vf);1167		mutex_unlock(&vf->status_lock);1168	}1169}1170 1171static void efx_siena_sriov_free_local(struct efx_nic *efx)1172{1173	struct siena_nic_data *nic_data = efx->nic_data;1174	struct efx_local_addr *local_addr;1175	struct efx_endpoint_page *epp;1176 1177	while (!list_empty(&nic_data->local_addr_list)) {1178		local_addr = list_first_entry(&nic_data->local_addr_list,1179					      struct efx_local_addr, link);1180		list_del(&local_addr->link);1181		kfree(local_addr);1182	}1183 1184	while (!list_empty(&nic_data->local_page_list)) {1185		epp = list_first_entry(&nic_data->local_page_list,1186				       struct efx_endpoint_page, link);1187		list_del(&epp->link);1188		dma_free_coherent(&efx->pci_dev->dev, EFX_PAGE_SIZE,1189				  epp->ptr, epp->addr);1190		kfree(epp);1191	}1192}1193 1194static int efx_siena_sriov_vf_alloc(struct efx_nic *efx)1195{1196	unsigned index;1197	struct siena_vf *vf;1198	struct siena_nic_data *nic_data = efx->nic_data;1199 1200	nic_data->vf = kcalloc(efx->vf_count, sizeof(*nic_data->vf),1201			       GFP_KERNEL);1202	if (!nic_data->vf)1203		return -ENOMEM;1204 1205	for (index = 0; index < efx->vf_count; ++index) {1206		vf = nic_data->vf + index;1207 1208		vf->efx = efx;1209		vf->index = index;1210		vf->rx_filter_id = -1;1211		vf->tx_filter_mode = VF_TX_FILTER_AUTO;1212		vf->tx_filter_id = -1;1213		INIT_WORK(&vf->req, efx_siena_sriov_vfdi);1214		INIT_WORK(&vf->reset_work, efx_siena_sriov_reset_vf_work);1215		init_waitqueue_head(&vf->flush_waitq);1216		mutex_init(&vf->status_lock);1217		mutex_init(&vf->txq_lock);1218	}1219 1220	return 0;1221}1222 1223static void efx_siena_sriov_vfs_fini(struct efx_nic *efx)1224{1225	struct siena_nic_data *nic_data = efx->nic_data;1226	struct siena_vf *vf;1227	unsigned int pos;1228 1229	for (pos = 0; pos < efx->vf_count; ++pos) {1230		vf = nic_data->vf + pos;1231 1232		efx_siena_free_buffer(efx, &vf->buf);1233		kfree(vf->peer_page_addrs);1234		vf->peer_page_addrs = NULL;1235		vf->peer_page_count = 0;1236 1237		vf->evq0_count = 0;1238	}1239}1240 1241static int efx_siena_sriov_vfs_init(struct efx_nic *efx)1242{1243	struct pci_dev *pci_dev = efx->pci_dev;1244	struct siena_nic_data *nic_data = efx->nic_data;1245	unsigned index, devfn, sriov, buftbl_base;1246	u16 offset, stride;1247	struct siena_vf *vf;1248	int rc;1249 1250	sriov = pci_find_ext_capability(pci_dev, PCI_EXT_CAP_ID_SRIOV);1251	if (!sriov)1252		return -ENOENT;1253 1254	pci_read_config_word(pci_dev, sriov + PCI_SRIOV_VF_OFFSET, &offset);1255	pci_read_config_word(pci_dev, sriov + PCI_SRIOV_VF_STRIDE, &stride);1256 1257	buftbl_base = nic_data->vf_buftbl_base;1258	devfn = pci_dev->devfn + offset;1259	for (index = 0; index < efx->vf_count; ++index) {1260		vf = nic_data->vf + index;1261 1262		/* Reserve buffer entries */1263		vf->buftbl_base = buftbl_base;1264		buftbl_base += EFX_VF_BUFTBL_PER_VI * efx_vf_size(efx);1265 1266		vf->pci_rid = devfn;1267		snprintf(vf->pci_name, sizeof(vf->pci_name),1268			 "%04x:%02x:%02x.%d",1269			 pci_domain_nr(pci_dev->bus), pci_dev->bus->number,1270			 PCI_SLOT(devfn), PCI_FUNC(devfn));1271 1272		rc = efx_siena_alloc_buffer(efx, &vf->buf, EFX_PAGE_SIZE,1273					    GFP_KERNEL);1274		if (rc)1275			goto fail;1276 1277		devfn += stride;1278	}1279 1280	return 0;1281 1282fail:1283	efx_siena_sriov_vfs_fini(efx);1284	return rc;1285}1286 1287int efx_siena_sriov_init(struct efx_nic *efx)1288{1289	struct net_device *net_dev = efx->net_dev;1290	struct siena_nic_data *nic_data = efx->nic_data;1291	struct vfdi_status *vfdi_status;1292	int rc;1293 1294	/* Ensure there's room for vf_channel */1295	BUILD_BUG_ON(EFX_MAX_CHANNELS + 1 >= EFX_VI_BASE);1296	/* Ensure that VI_BASE is aligned on VI_SCALE */1297	BUILD_BUG_ON(EFX_VI_BASE & ((1 << EFX_VI_SCALE_MAX) - 1));1298 1299	if (efx->vf_count == 0)1300		return 0;1301 1302	rc = efx_siena_sriov_cmd(efx, true, NULL, NULL);1303	if (rc)1304		goto fail_cmd;1305 1306	rc = efx_siena_alloc_buffer(efx, &nic_data->vfdi_status,1307				    sizeof(*vfdi_status), GFP_KERNEL);1308	if (rc)1309		goto fail_status;1310	vfdi_status = nic_data->vfdi_status.addr;1311	memset(vfdi_status, 0, sizeof(*vfdi_status));1312	vfdi_status->version = 1;1313	vfdi_status->length = sizeof(*vfdi_status);1314	vfdi_status->max_tx_channels = vf_max_tx_channels;1315	vfdi_status->vi_scale = efx->vi_scale;1316	vfdi_status->rss_rxq_count = efx->rss_spread;1317	vfdi_status->peer_count = 1 + efx->vf_count;1318	vfdi_status->timer_quantum_ns = efx->timer_quantum_ns;1319 1320	rc = efx_siena_sriov_vf_alloc(efx);1321	if (rc)1322		goto fail_alloc;1323 1324	mutex_init(&nic_data->local_lock);1325	INIT_WORK(&nic_data->peer_work, efx_siena_sriov_peer_work);1326	INIT_LIST_HEAD(&nic_data->local_addr_list);1327	INIT_LIST_HEAD(&nic_data->local_page_list);1328 1329	rc = efx_siena_sriov_vfs_init(efx);1330	if (rc)1331		goto fail_vfs;1332 1333	rtnl_lock();1334	ether_addr_copy(vfdi_status->peers[0].mac_addr, net_dev->dev_addr);1335	efx->vf_init_count = efx->vf_count;1336	rtnl_unlock();1337 1338	efx_siena_sriov_usrev(efx, true);1339 1340	/* At this point we must be ready to accept VFDI requests */1341 1342	rc = pci_enable_sriov(efx->pci_dev, efx->vf_count);1343	if (rc)1344		goto fail_pci;1345 1346	netif_info(efx, probe, net_dev,1347		   "enabled SR-IOV for %d VFs, %d VI per VF\n",1348		   efx->vf_count, efx_vf_size(efx));1349	return 0;1350 1351fail_pci:1352	efx_siena_sriov_usrev(efx, false);1353	rtnl_lock();1354	efx->vf_init_count = 0;1355	rtnl_unlock();1356	efx_siena_sriov_vfs_fini(efx);1357fail_vfs:1358	cancel_work_sync(&nic_data->peer_work);1359	efx_siena_sriov_free_local(efx);1360	kfree(nic_data->vf);1361fail_alloc:1362	efx_siena_free_buffer(efx, &nic_data->vfdi_status);1363fail_status:1364	efx_siena_sriov_cmd(efx, false, NULL, NULL);1365fail_cmd:1366	return rc;1367}1368 1369void efx_siena_sriov_fini(struct efx_nic *efx)1370{1371	struct siena_vf *vf;1372	unsigned int pos;1373	struct siena_nic_data *nic_data = efx->nic_data;1374 1375	if (efx->vf_init_count == 0)1376		return;1377 1378	/* Disable all interfaces to reconfiguration */1379	BUG_ON(nic_data->vfdi_channel->enabled);1380	efx_siena_sriov_usrev(efx, false);1381	rtnl_lock();1382	efx->vf_init_count = 0;1383	rtnl_unlock();1384 1385	/* Flush all reconfiguration work */1386	for (pos = 0; pos < efx->vf_count; ++pos) {1387		vf = nic_data->vf + pos;1388		cancel_work_sync(&vf->req);1389		cancel_work_sync(&vf->reset_work);1390	}1391	cancel_work_sync(&nic_data->peer_work);1392 1393	pci_disable_sriov(efx->pci_dev);1394 1395	/* Tear down back-end state */1396	efx_siena_sriov_vfs_fini(efx);1397	efx_siena_sriov_free_local(efx);1398	kfree(nic_data->vf);1399	efx_siena_free_buffer(efx, &nic_data->vfdi_status);1400	efx_siena_sriov_cmd(efx, false, NULL, NULL);1401}1402 1403void efx_siena_sriov_event(struct efx_channel *channel, efx_qword_t *event)1404{1405	struct efx_nic *efx = channel->efx;1406	struct siena_vf *vf;1407	unsigned qid, seq, type, data;1408 1409	qid = EFX_QWORD_FIELD(*event, FSF_CZ_USER_QID);1410 1411	/* USR_EV_REG_VALUE is dword0, so access the VFDI_EV fields directly */1412	BUILD_BUG_ON(FSF_CZ_USER_EV_REG_VALUE_LBN != 0);1413	seq = EFX_QWORD_FIELD(*event, VFDI_EV_SEQ);1414	type = EFX_QWORD_FIELD(*event, VFDI_EV_TYPE);1415	data = EFX_QWORD_FIELD(*event, VFDI_EV_DATA);1416 1417	netif_vdbg(efx, hw, efx->net_dev,1418		   "USR_EV event from qid %d seq 0x%x type %d data 0x%x\n",1419		   qid, seq, type, data);1420 1421	if (map_vi_index(efx, qid, &vf, NULL))1422		return;1423	if (vf->busy)1424		goto error;1425 1426	if (type == VFDI_EV_TYPE_REQ_WORD0) {1427		/* Resynchronise */1428		vf->req_type = VFDI_EV_TYPE_REQ_WORD0;1429		vf->req_seqno = seq + 1;1430		vf->req_addr = 0;1431	} else if (seq != (vf->req_seqno++ & 0xff) || type != vf->req_type)1432		goto error;1433 1434	switch (vf->req_type) {1435	case VFDI_EV_TYPE_REQ_WORD0:1436	case VFDI_EV_TYPE_REQ_WORD1:1437	case VFDI_EV_TYPE_REQ_WORD2:1438		vf->req_addr |= (u64)data << (vf->req_type << 4);1439		++vf->req_type;1440		return;1441 1442	case VFDI_EV_TYPE_REQ_WORD3:1443		vf->req_addr |= (u64)data << 48;1444		vf->req_type = VFDI_EV_TYPE_REQ_WORD0;1445		vf->busy = true;1446		queue_work(vfdi_workqueue, &vf->req);1447		return;1448	}1449 1450error:1451	if (net_ratelimit())1452		netif_err(efx, hw, efx->net_dev,1453			  "ERROR: Screaming VFDI request from %s\n",1454			  vf->pci_name);1455	/* Reset the request and sequence number */1456	vf->req_type = VFDI_EV_TYPE_REQ_WORD0;1457	vf->req_seqno = seq + 1;1458}1459 1460void efx_siena_sriov_flr(struct efx_nic *efx, unsigned vf_i)1461{1462	struct siena_nic_data *nic_data = efx->nic_data;1463	struct siena_vf *vf;1464 1465	if (vf_i > efx->vf_init_count)1466		return;1467	vf = nic_data->vf + vf_i;1468	netif_info(efx, hw, efx->net_dev,1469		   "FLR on VF %s\n", vf->pci_name);1470 1471	vf->status_addr = 0;1472	efx_vfdi_remove_all_filters(vf);1473	efx_vfdi_flush_clear(vf);1474 1475	vf->evq0_count = 0;1476}1477 1478int efx_siena_sriov_mac_address_changed(struct efx_nic *efx)1479{1480	struct siena_nic_data *nic_data = efx->nic_data;1481	struct vfdi_status *vfdi_status = nic_data->vfdi_status.addr;1482 1483	if (!efx->vf_init_count)1484		return 0;1485	ether_addr_copy(vfdi_status->peers[0].mac_addr,1486			efx->net_dev->dev_addr);1487	queue_work(vfdi_workqueue, &nic_data->peer_work);1488 1489	return 0;1490}1491 1492void efx_siena_sriov_tx_flush_done(struct efx_nic *efx, efx_qword_t *event)1493{1494	struct siena_vf *vf;1495	unsigned queue, qid;1496 1497	queue = EFX_QWORD_FIELD(*event,  FSF_AZ_DRIVER_EV_SUBDATA);1498	if (map_vi_index(efx, queue, &vf, &qid))1499		return;1500	/* Ignore flush completions triggered by an FLR */1501	if (!test_bit(qid, vf->txq_mask))1502		return;1503 1504	__clear_bit(qid, vf->txq_mask);1505	--vf->txq_count;1506 1507	if (efx_vfdi_flush_wake(vf))1508		wake_up(&vf->flush_waitq);1509}1510 1511void efx_siena_sriov_rx_flush_done(struct efx_nic *efx, efx_qword_t *event)1512{1513	struct siena_vf *vf;1514	unsigned ev_failed, queue, qid;1515 1516	queue = EFX_QWORD_FIELD(*event, FSF_AZ_DRIVER_EV_RX_DESCQ_ID);1517	ev_failed = EFX_QWORD_FIELD(*event,1518				    FSF_AZ_DRIVER_EV_RX_FLUSH_FAIL);1519	if (map_vi_index(efx, queue, &vf, &qid))1520		return;1521	if (!test_bit(qid, vf->rxq_mask))1522		return;1523 1524	if (ev_failed) {1525		set_bit(qid, vf->rxq_retry_mask);1526		atomic_inc(&vf->rxq_retry_count);1527	} else {1528		__clear_bit(qid, vf->rxq_mask);1529		--vf->rxq_count;1530	}1531	if (efx_vfdi_flush_wake(vf))1532		wake_up(&vf->flush_waitq);1533}1534 1535/* Called from napi. Schedule the reset work item */1536void efx_siena_sriov_desc_fetch_err(struct efx_nic *efx, unsigned dmaq)1537{1538	struct siena_vf *vf;1539	unsigned int rel;1540 1541	if (map_vi_index(efx, dmaq, &vf, &rel))1542		return;1543 1544	if (net_ratelimit())1545		netif_err(efx, hw, efx->net_dev,1546			  "VF %d DMA Q %d reports descriptor fetch error.\n",1547			  vf->index, rel);1548	queue_work(vfdi_workqueue, &vf->reset_work);1549}1550 1551/* Reset all VFs */1552void efx_siena_sriov_reset(struct efx_nic *efx)1553{1554	struct siena_nic_data *nic_data = efx->nic_data;1555	unsigned int vf_i;1556	struct efx_buffer buf;1557	struct siena_vf *vf;1558 1559	ASSERT_RTNL();1560 1561	if (efx->vf_init_count == 0)1562		return;1563 1564	efx_siena_sriov_usrev(efx, true);1565	(void)efx_siena_sriov_cmd(efx, true, NULL, NULL);1566 1567	if (efx_siena_alloc_buffer(efx, &buf, EFX_PAGE_SIZE, GFP_NOIO))1568		return;1569 1570	for (vf_i = 0; vf_i < efx->vf_init_count; ++vf_i) {1571		vf = nic_data->vf + vf_i;1572		efx_siena_sriov_reset_vf(vf, &buf);1573	}1574 1575	efx_siena_free_buffer(efx, &buf);1576}1577 1578int efx_init_sriov(void)1579{1580	/* A single threaded workqueue is sufficient. efx_siena_sriov_vfdi() and1581	 * efx_siena_sriov_peer_work() spend almost all their time sleeping for1582	 * MCDI to complete anyway1583	 */1584	vfdi_workqueue = create_singlethread_workqueue("sfc_vfdi");1585	if (!vfdi_workqueue)1586		return -ENOMEM;1587	return 0;1588}1589 1590void efx_fini_sriov(void)1591{1592	destroy_workqueue(vfdi_workqueue);1593}1594 1595int efx_siena_sriov_set_vf_mac(struct efx_nic *efx, int vf_i, const u8 *mac)1596{1597	struct siena_nic_data *nic_data = efx->nic_data;1598	struct siena_vf *vf;1599 1600	if (vf_i >= efx->vf_init_count)1601		return -EINVAL;1602	vf = nic_data->vf + vf_i;1603 1604	mutex_lock(&vf->status_lock);1605	ether_addr_copy(vf->addr.mac_addr, mac);1606	__efx_siena_sriov_update_vf_addr(vf);1607	mutex_unlock(&vf->status_lock);1608 1609	return 0;1610}1611 1612int efx_siena_sriov_set_vf_vlan(struct efx_nic *efx, int vf_i,1613				u16 vlan, u8 qos)1614{1615	struct siena_nic_data *nic_data = efx->nic_data;1616	struct siena_vf *vf;1617	u16 tci;1618 1619	if (vf_i >= efx->vf_init_count)1620		return -EINVAL;1621	vf = nic_data->vf + vf_i;1622 1623	mutex_lock(&vf->status_lock);1624	tci = (vlan & VLAN_VID_MASK) | ((qos & 0x7) << VLAN_PRIO_SHIFT);1625	vf->addr.tci = htons(tci);1626	__efx_siena_sriov_update_vf_addr(vf);1627	mutex_unlock(&vf->status_lock);1628 1629	return 0;1630}1631 1632int efx_siena_sriov_set_vf_spoofchk(struct efx_nic *efx, int vf_i,1633				    bool spoofchk)1634{1635	struct siena_nic_data *nic_data = efx->nic_data;1636	struct siena_vf *vf;1637	int rc;1638 1639	if (vf_i >= efx->vf_init_count)1640		return -EINVAL;1641	vf = nic_data->vf + vf_i;1642 1643	mutex_lock(&vf->txq_lock);1644	if (vf->txq_count == 0) {1645		vf->tx_filter_mode =1646			spoofchk ? VF_TX_FILTER_ON : VF_TX_FILTER_OFF;1647		rc = 0;1648	} else {1649		/* This cannot be changed while TX queues are running */1650		rc = -EBUSY;1651	}1652	mutex_unlock(&vf->txq_lock);1653	return rc;1654}1655 1656int efx_siena_sriov_get_vf_config(struct efx_nic *efx, int vf_i,1657				  struct ifla_vf_info *ivi)1658{1659	struct siena_nic_data *nic_data = efx->nic_data;1660	struct siena_vf *vf;1661	u16 tci;1662 1663	if (vf_i >= efx->vf_init_count)1664		return -EINVAL;1665	vf = nic_data->vf + vf_i;1666 1667	ivi->vf = vf_i;1668	ether_addr_copy(ivi->mac, vf->addr.mac_addr);1669	ivi->max_tx_rate = 0;1670	ivi->min_tx_rate = 0;1671	tci = ntohs(vf->addr.tci);1672	ivi->vlan = tci & VLAN_VID_MASK;1673	ivi->qos = (tci >> VLAN_PRIO_SHIFT) & 0x7;1674	ivi->spoofchk = vf->tx_filter_mode == VF_TX_FILTER_ON;1675 1676	return 0;1677}1678 1679bool efx_siena_sriov_wanted(struct efx_nic *efx)1680{1681	return efx->vf_count != 0;1682}1683 1684int efx_siena_sriov_configure(struct efx_nic *efx, int num_vfs)1685{1686	return 0;1687}1688