~hc/RK356X_SDK_RELEASE.git

..	..	@@ -1,33 +1,6 @@
	1	+// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
1	2	/*
2		- * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
3		- *
4		- * This software is available to you under a choice of one of two
5		- * licenses. You may choose to be licensed under the terms of the GNU
6		- * General Public License (GPL) Version 2, available from the file
7		- * COPYING in the main directory of this source tree, or the
8		- * OpenIB.org BSD license below:
9		- *
10		- * Redistribution and use in source and binary forms, with or
11		- * without modification, are permitted provided that the following
12		- * conditions are met:
13		- *
14		- * - Redistributions of source code must retain the above
15		- * copyright notice, this list of conditions and the following
16		- * disclaimer.
17		- *
18		- * - Redistributions in binary form must reproduce the above
19		- * copyright notice, this list of conditions and the following
20		- * disclaimer in the documentation and/or other materials
21		- * provided with the distribution.
22		- *
23		- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24		- * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25		- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26		- * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27		- * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28		- * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29		- * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30		- * SOFTWARE.
	3	+ * Copyright (c) 2013-2020, Mellanox Technologies inc. All rights reserved.
31	4	*/
32	5
33	6	#include <linux/debugfs.h>
..	..	@@ -39,9 +12,6 @@
39	12	#include <linux/dma-mapping.h>
40	13	#include <linux/slab.h>
41	14	#include <linux/bitmap.h>
42		-#if defined(CONFIG_X86)
43		-#include <asm/pat.h>
44		-#endif
45	15	#include <linux/sched.h>
46	16	#include <linux/sched/mm.h>
47	17	#include <linux/sched/task.h>
..	..	@@ -52,40 +22,45 @@
52	22	#include <linux/mlx5/port.h>
53	23	#include <linux/mlx5/vport.h>
54	24	#include <linux/mlx5/fs.h>
	25	+#include <linux/mlx5/eswitch.h>
55	26	#include <linux/list.h>
56	27	#include <rdma/ib_smi.h>
57	28	#include <rdma/ib_umem.h>
	29	+#include <rdma/lag.h>
58	30	#include <linux/in.h>
59	31	#include <linux/etherdevice.h>
60	32	#include "mlx5_ib.h"
61	33	#include "ib_rep.h"
62	34	#include "cmd.h"
63		-#include <linux/mlx5/fs_helpers.h>
	35	+#include "devx.h"
	36	+#include "fs.h"
	37	+#include "srq.h"
	38	+#include "qp.h"
	39	+#include "wr.h"
	40	+#include "restrack.h"
	41	+#include "counters.h"
64	42	#include <linux/mlx5/accel.h>
65	43	#include <rdma/uverbs_std_types.h>
66	44	#include <rdma/mlx5_user_ioctl_verbs.h>
67	45	#include <rdma/mlx5_user_ioctl_cmds.h>
	46	+#include <rdma/ib_umem_odp.h>
68	47
69	48	#define UVERBS_MODULE_NAME mlx5_ib
70	49	#include <rdma/uverbs_named_ioctl.h>
71	50
72		-#define DRIVER_NAME "mlx5_ib"
73		-#define DRIVER_VERSION "5.0-0"
74		-
75	51	MODULE_AUTHOR("Eli Cohen <eli@mellanox.com>");
76		-MODULE_DESCRIPTION("Mellanox Connect-IB HCA IB driver");
	52	+MODULE_DESCRIPTION("Mellanox 5th generation network adapters (ConnectX series) IB driver");
77	53	MODULE_LICENSE("Dual BSD/GPL");
78		-
79		-static char mlx5_version[] =
80		- DRIVER_NAME ": Mellanox Connect-IB Infiniband driver v"
81		- DRIVER_VERSION "\n";
82	54
83	55	struct mlx5_ib_event_work {
84	56	struct work_struct work;
85		- struct mlx5_core_dev *dev;
86		- void *context;
87		- enum mlx5_dev_event event;
88		- unsigned long param;
	57	+ union {
	58	+ struct mlx5_ib_dev *dev;
	59	+ struct mlx5_ib_multiport_info *mpi;
	60	+ };
	61	+ bool is_slave;
	62	+ unsigned int event;
	63	+ void *param;
89	64	};
90	65
91	66	enum {
..	..	@@ -146,10 +121,38 @@
146	121	int ret;
147	122
148	123	memset(&attr, 0, sizeof(attr));
149		- ret = ibdev->query_port(ibdev, port_num, &attr);
	124	+ ret = ibdev->ops.query_port(ibdev, port_num, &attr);
150	125	if (!ret)
151	126	*state = attr.state;
152	127	return ret;
	128	+}
	129	+
	130	+static struct mlx5_roce mlx5_get_rep_roce(struct mlx5_ib_dev dev,
	131	+ struct net_device *ndev,
	132	+ u8 *port_num)
	133	+{
	134	+ struct mlx5_eswitch *esw = dev->mdev->priv.eswitch;
	135	+ struct net_device *rep_ndev;
	136	+ struct mlx5_ib_port *port;
	137	+ int i;
	138	+
	139	+ for (i = 0; i < dev->num_ports; i++) {
	140	+ port = &dev->port[i];
	141	+ if (!port->rep)
	142	+ continue;
	143	+
	144	+ read_lock(&port->roce.netdev_lock);
	145	+ rep_ndev = mlx5_ib_get_rep_netdev(esw,
	146	+ port->rep->vport);
	147	+ if (rep_ndev == ndev) {
	148	+ read_unlock(&port->roce.netdev_lock);
	149	+ *port_num = i + 1;
	150	+ return &port->roce;
	151	+ }
	152	+ read_unlock(&port->roce.netdev_lock);
	153	+ }
	154	+
	155	+ return NULL;
153	156	}
154	157
155	158	static int mlx5_netdev_event(struct notifier_block *this,
..	..	@@ -168,21 +171,20 @@
168	171
169	172	switch (event) {
170	173	case NETDEV_REGISTER:
171		- case NETDEV_UNREGISTER:
	174	+ /* Should already be registered during the load */
	175	+ if (ibdev->is_rep)
	176	+ break;
172	177	write_lock(&roce->netdev_lock);
173		- if (ibdev->rep) {
174		- struct mlx5_eswitch *esw = ibdev->mdev->priv.eswitch;
175		- struct net_device *rep_ndev;
	178	+ if (ndev->dev.parent == mdev->device)
	179	+ roce->netdev = ndev;
	180	+ write_unlock(&roce->netdev_lock);
	181	+ break;
176	182
177		- rep_ndev = mlx5_ib_get_rep_netdev(esw,
178		- ibdev->rep->vport);
179		- if (rep_ndev == ndev)
180		- roce->netdev = (event == NETDEV_UNREGISTER) ?
181		- NULL : ndev;
182		- } else if (ndev->dev.parent == &mdev->pdev->dev) {
183		- roce->netdev = (event == NETDEV_UNREGISTER) ?
184		- NULL : ndev;
185		- }
	183	+ case NETDEV_UNREGISTER:
	184	+ /* In case of reps, ib device goes away before the netdevs */
	185	+ write_lock(&roce->netdev_lock);
	186	+ if (roce->netdev == ndev)
	187	+ roce->netdev = NULL;
186	188	write_unlock(&roce->netdev_lock);
187	189	break;
188	190
..	..	@@ -197,6 +199,10 @@
197	199	dev_put(lag_ndev);
198	200	}
199	201
	202	+ if (ibdev->is_rep)
	203	+ roce = mlx5_get_rep_roce(ibdev, ndev, &port_num);
	204	+ if (!roce)
	205	+ return NOTIFY_DONE;
200	206	if ((upper == ndev \|\| (!upper && ndev == roce->netdev))
201	207	&& ibdev->ib_active) {
202	208	struct ib_event ibev = { };
..	..	@@ -249,11 +255,11 @@
249	255
250	256	/* Ensure ndev does not disappear before we invoke dev_hold()
251	257	*/
252		- read_lock(&ibdev->roce[port_num - 1].netdev_lock);
253		- ndev = ibdev->roce[port_num - 1].netdev;
	258	+ read_lock(&ibdev->port[port_num - 1].roce.netdev_lock);
	259	+ ndev = ibdev->port[port_num - 1].roce.netdev;
254	260	if (ndev)
255	261	dev_hold(ndev);
256		- read_unlock(&ibdev->roce[port_num - 1].netdev_lock);
	262	+ read_unlock(&ibdev->port[port_num - 1].roce.netdev_lock);
257	263
258	264	out:
259	265	mlx5_ib_put_native_port_mdev(ibdev, port_num);
..	..	@@ -281,9 +287,6 @@
281	287	*native_port_num = 1;
282	288
283	289	port = &ibdev->port[ib_port_num - 1];
284		- if (!port)
285		- return NULL;
286		-
287	290	spin_lock(&port->mp.mpi_lock);
288	291	mpi = ibdev->port[ib_port_num - 1].mp.mpi;
289	292	if (mpi && !mpi->unaffiliate) {
..	..	@@ -323,8 +326,8 @@
323	326	spin_unlock(&port->mp.mpi_lock);
324	327	}
325	328
326		-static int translate_eth_proto_oper(u32 eth_proto_oper, u8 *active_speed,
327		- u8 *active_width)
	329	+static int translate_eth_legacy_proto_oper(u32 eth_proto_oper,
	330	+ u16 active_speed, u8 active_width)
328	331	{
329	332	switch (eth_proto_oper) {
330	333	case MLX5E_PROT_MASK(MLX5E_1000BASE_CX_SGMII):
..	..	@@ -381,10 +384,89 @@
381	384	return 0;
382	385	}
383	386
	387	+static int translate_eth_ext_proto_oper(u32 eth_proto_oper, u16 *active_speed,
	388	+ u8 *active_width)
	389	+{
	390	+ switch (eth_proto_oper) {
	391	+ case MLX5E_PROT_MASK(MLX5E_SGMII_100M):
	392	+ case MLX5E_PROT_MASK(MLX5E_1000BASE_X_SGMII):
	393	+ *active_width = IB_WIDTH_1X;
	394	+ *active_speed = IB_SPEED_SDR;
	395	+ break;
	396	+ case MLX5E_PROT_MASK(MLX5E_5GBASE_R):
	397	+ *active_width = IB_WIDTH_1X;
	398	+ *active_speed = IB_SPEED_DDR;
	399	+ break;
	400	+ case MLX5E_PROT_MASK(MLX5E_10GBASE_XFI_XAUI_1):
	401	+ *active_width = IB_WIDTH_1X;
	402	+ *active_speed = IB_SPEED_QDR;
	403	+ break;
	404	+ case MLX5E_PROT_MASK(MLX5E_40GBASE_XLAUI_4_XLPPI_4):
	405	+ *active_width = IB_WIDTH_4X;
	406	+ *active_speed = IB_SPEED_QDR;
	407	+ break;
	408	+ case MLX5E_PROT_MASK(MLX5E_25GAUI_1_25GBASE_CR_KR):
	409	+ *active_width = IB_WIDTH_1X;
	410	+ *active_speed = IB_SPEED_EDR;
	411	+ break;
	412	+ case MLX5E_PROT_MASK(MLX5E_50GAUI_2_LAUI_2_50GBASE_CR2_KR2):
	413	+ *active_width = IB_WIDTH_2X;
	414	+ *active_speed = IB_SPEED_EDR;
	415	+ break;
	416	+ case MLX5E_PROT_MASK(MLX5E_50GAUI_1_LAUI_1_50GBASE_CR_KR):
	417	+ *active_width = IB_WIDTH_1X;
	418	+ *active_speed = IB_SPEED_HDR;
	419	+ break;
	420	+ case MLX5E_PROT_MASK(MLX5E_CAUI_4_100GBASE_CR4_KR4):
	421	+ *active_width = IB_WIDTH_4X;
	422	+ *active_speed = IB_SPEED_EDR;
	423	+ break;
	424	+ case MLX5E_PROT_MASK(MLX5E_100GAUI_2_100GBASE_CR2_KR2):
	425	+ *active_width = IB_WIDTH_2X;
	426	+ *active_speed = IB_SPEED_HDR;
	427	+ break;
	428	+ case MLX5E_PROT_MASK(MLX5E_100GAUI_1_100GBASE_CR_KR):
	429	+ *active_width = IB_WIDTH_1X;
	430	+ *active_speed = IB_SPEED_NDR;
	431	+ break;
	432	+ case MLX5E_PROT_MASK(MLX5E_200GAUI_4_200GBASE_CR4_KR4):
	433	+ *active_width = IB_WIDTH_4X;
	434	+ *active_speed = IB_SPEED_HDR;
	435	+ break;
	436	+ case MLX5E_PROT_MASK(MLX5E_200GAUI_2_200GBASE_CR2_KR2):
	437	+ *active_width = IB_WIDTH_2X;
	438	+ *active_speed = IB_SPEED_NDR;
	439	+ break;
	440	+ case MLX5E_PROT_MASK(MLX5E_400GAUI_8):
	441	+ *active_width = IB_WIDTH_8X;
	442	+ *active_speed = IB_SPEED_HDR;
	443	+ break;
	444	+ case MLX5E_PROT_MASK(MLX5E_400GAUI_4_400GBASE_CR4_KR4):
	445	+ *active_width = IB_WIDTH_4X;
	446	+ *active_speed = IB_SPEED_NDR;
	447	+ break;
	448	+ default:
	449	+ return -EINVAL;
	450	+ }
	451	+
	452	+ return 0;
	453	+}
	454	+
	455	+static int translate_eth_proto_oper(u32 eth_proto_oper, u16 *active_speed,
	456	+ u8 *active_width, bool ext)
	457	+{
	458	+ return ext ?
	459	+ translate_eth_ext_proto_oper(eth_proto_oper, active_speed,
	460	+ active_width) :
	461	+ translate_eth_legacy_proto_oper(eth_proto_oper, active_speed,
	462	+ active_width);
	463	+}
	464	+
384	465	static int mlx5_query_port_roce(struct ib_device *device, u8 port_num,
385	466	struct ib_port_attr *props)
386	467	{
387	468	struct mlx5_ib_dev *dev = to_mdev(device);
	469	+ u32 out[MLX5_ST_SZ_DW(ptys_reg)] = {0};
388	470	struct mlx5_core_dev *mdev;
389	471	struct net_device ndev, upper;
390	472	enum ib_mtu ndev_ib_mtu;
..	..	@@ -392,6 +474,7 @@
392	474	u16 qkey_viol_cntr;
393	475	u32 eth_prot_oper;
394	476	u8 mdev_port_num;
	477	+ bool ext;
395	478	int err;
396	479
397	480	mdev = mlx5_ib_get_native_port_mdev(dev, port_num, &mdev_port_num);
..	..	@@ -407,17 +490,24 @@
407	490
408	491	/* Possible bad flows are checked before filling out props so in case
409	492	* of an error it will still be zeroed out.
	493	+ * Use native port in case of reps
410	494	*/
411		- err = mlx5_query_port_eth_proto_oper(mdev, &eth_prot_oper,
412		- mdev_port_num);
	495	+ if (dev->is_rep)
	496	+ err = mlx5_query_port_ptys(mdev, out, sizeof(out), MLX5_PTYS_EN,
	497	+ 1);
	498	+ else
	499	+ err = mlx5_query_port_ptys(mdev, out, sizeof(out), MLX5_PTYS_EN,
	500	+ mdev_port_num);
413	501	if (err)
414	502	goto out;
	503	+ ext = !!MLX5_GET_ETH_PROTO(ptys_reg, out, true, eth_proto_capability);
	504	+ eth_prot_oper = MLX5_GET_ETH_PROTO(ptys_reg, out, ext, eth_proto_oper);
415	505
416	506	props->active_width = IB_WIDTH_4X;
417	507	props->active_speed = IB_SPEED_QDR;
418	508
419	509	translate_eth_proto_oper(eth_prot_oper, &props->active_speed,
420		- &props->active_width);
	510	+ &props->active_width, ext);
421	511
422	512	props->port_cap_flags \|= IB_PORT_CM_SUP;
423	513	props->ip_gids = true;
..	..	@@ -428,7 +518,7 @@
428	518	props->max_msg_sz = 1 << MLX5_CAP_GEN(dev->mdev, log_max_msg);
429	519	props->pkey_tbl_len = 1;
430	520	props->state = IB_PORT_DOWN;
431		- props->phys_state = 3;
	521	+ props->phys_state = IB_PORT_PHYS_STATE_DISABLED;
432	522
433	523	mlx5_query_nic_vport_qkey_viol_cntr(mdev, &qkey_viol_cntr);
434	524	props->qkey_viol_cntr = qkey_viol_cntr;
..	..	@@ -441,7 +531,7 @@
441	531	if (!ndev)
442	532	goto out;
443	533
444		- if (mlx5_lag_is_active(dev->mdev)) {
	534	+ if (dev->lag_active) {
445	535	rcu_read_lock();
446	536	upper = netdev_master_upper_dev_get_rcu(ndev);
447	537	if (upper) {
..	..	@@ -454,7 +544,7 @@
454	544
455	545	if (netif_running(ndev) && netif_carrier_ok(ndev)) {
456	546	props->state = IB_PORT_ACTIVE;
457		- props->phys_state = 5;
	547	+ props->phys_state = IB_PORT_PHYS_STATE_LINK_UP;
458	548	}
459	549
460	550	ndev_ib_mtu = iboe_get_mtu(ndev->mtu);
..	..	@@ -472,25 +562,22 @@
472	562	unsigned int index, const union ib_gid *gid,
473	563	const struct ib_gid_attr *attr)
474	564	{
475		- enum ib_gid_type gid_type = IB_GID_TYPE_IB;
	565	+ enum ib_gid_type gid_type = IB_GID_TYPE_ROCE;
	566	+ u16 vlan_id = 0xffff;
476	567	u8 roce_version = 0;
477	568	u8 roce_l3_type = 0;
478		- bool vlan = false;
479	569	u8 mac[ETH_ALEN];
480		- u16 vlan_id = 0;
	570	+ int ret;
481	571
482	572	if (gid) {
483	573	gid_type = attr->gid_type;
484		- ether_addr_copy(mac, attr->ndev->dev_addr);
485		-
486		- if (is_vlan_dev(attr->ndev)) {
487		- vlan = true;
488		- vlan_id = vlan_dev_vlan_id(attr->ndev);
489		- }
	574	+ ret = rdma_read_gid_l2_fields(attr, &vlan_id, &mac[0]);
	575	+ if (ret)
	576	+ return ret;
490	577	}
491	578
492	579	switch (gid_type) {
493		- case IB_GID_TYPE_IB:
	580	+ case IB_GID_TYPE_ROCE:
494	581	roce_version = MLX5_ROCE_VERSION_1;
495	582	break;
496	583	case IB_GID_TYPE_ROCE_UDP_ENCAP:
..	..	@@ -506,8 +593,9 @@
506	593	}
507	594
508	595	return mlx5_core_roce_gid_set(dev->mdev, index, roce_version,
509		- roce_l3_type, gid->raw, mac, vlan,
510		- vlan_id, port_num);
	596	+ roce_l3_type, gid->raw, mac,
	597	+ vlan_id < VLAN_CFI_MASK, vlan_id,
	598	+ port_num);
511	599	}
512	600
513	601	static int mlx5_ib_add_gid(const struct ib_gid_attr *attr,
..	..	@@ -524,8 +612,8 @@
524	612	attr->index, NULL, NULL);
525	613	}
526	614
527		-__be16 mlx5_get_roce_udp_sport(struct mlx5_ib_dev *dev,
528		- const struct ib_gid_attr *attr)
	615	+__be16 mlx5_get_roce_udp_sport_min(const struct mlx5_ib_dev *dev,
	616	+ const struct ib_gid_attr *attr)
529	617	{
530	618	if (attr->gid_type != IB_GID_TYPE_ROCE_UDP_ENCAP)
531	619	return 0;
..	..	@@ -588,21 +676,6 @@
588	676	get_atomic_caps(dev, atomic_size_qp, props);
589	677	}
590	678
591		-static void get_atomic_caps_dc(struct mlx5_ib_dev *dev,
592		- struct ib_device_attr *props)
593		-{
594		- u8 atomic_size_qp = MLX5_CAP_ATOMIC(dev->mdev, atomic_size_dc);
595		-
596		- get_atomic_caps(dev, atomic_size_qp, props);
597		-}
598		-
599		-bool mlx5_ib_dc_atomic_is_supported(struct mlx5_ib_dev *dev)
600		-{
601		- struct ib_device_attr props = {};
602		-
603		- get_atomic_caps_dc(dev, &props);
604		- return (props.atomic_cap == IB_ATOMIC_HCA) ? true : false;
605		-}
606	679	static int mlx5_query_system_image_guid(struct ib_device *ibdev,
607	680	__be64 *sys_image_guid)
608	681	{
..	..	@@ -724,6 +797,7 @@
724	797	struct ib_device_attr *props,
725	798	struct ib_udata *uhw)
726	799	{
	800	+ size_t uhw_outlen = (uhw) ? uhw->outlen : 0;
727	801	struct mlx5_ib_dev *dev = to_mdev(ibdev);
728	802	struct mlx5_core_dev *mdev = dev->mdev;
729	803	int err = -ENOMEM;
..	..	@@ -737,12 +811,12 @@
737	811	u64 max_tso;
738	812
739	813	resp_len = sizeof(resp.comp_mask) + sizeof(resp.response_length);
740		- if (uhw->outlen && uhw->outlen < resp_len)
	814	+ if (uhw_outlen && uhw_outlen < resp_len)
741	815	return -EINVAL;
742		- else
743		- resp.response_length = resp_len;
744	816
745		- if (uhw->inlen && !ib_is_udata_cleared(uhw, 0, uhw->inlen))
	817	+ resp.response_length = resp_len;
	818	+
	819	+ if (uhw && uhw->inlen && !ib_is_udata_cleared(uhw, 0, uhw->inlen))
746	820	return -EINVAL;
747	821
748	822	memset(props, 0, sizeof(*props));
..	..	@@ -782,9 +856,11 @@
782	856	/* We support 'Gappy' memory registration too */
783	857	props->device_cap_flags \|= IB_DEVICE_SG_GAPS_REG;
784	858	}
785		- props->device_cap_flags \|= IB_DEVICE_MEM_MGT_EXTENSIONS;
	859	+ /* IB_WR_REG_MR always requires changing the entity size with UMR */
	860	+ if (!MLX5_CAP_GEN(dev->mdev, umr_modify_entity_size_disabled))
	861	+ props->device_cap_flags \|= IB_DEVICE_MEM_MGT_EXTENSIONS;
786	862	if (MLX5_CAP_GEN(mdev, sho)) {
787		- props->device_cap_flags \|= IB_DEVICE_SIGNATURE_HANDOVER;
	863	+ props->device_cap_flags \|= IB_DEVICE_INTEGRITY_HANDOVER;
788	864	/* At this stage no support for signature handover */
789	865	props->sig_prot_cap = IB_PROT_T10DIF_TYPE_1 \|
790	866	IB_PROT_T10DIF_TYPE_2 \|
..	..	@@ -806,7 +882,7 @@
806	882	props->raw_packet_caps \|=
807	883	IB_RAW_PACKET_CAP_CVLAN_STRIPPING;
808	884
809		- if (field_avail(typeof(resp), tso_caps, uhw->outlen)) {
	885	+ if (offsetofend(typeof(resp), tso_caps) <= uhw_outlen) {
810	886	max_tso = MLX5_CAP_ETH(mdev, max_lso_cap);
811	887	if (max_tso) {
812	888	resp.tso_caps.max_tso = 1 << max_tso;
..	..	@@ -816,7 +892,7 @@
816	892	}
817	893	}
818	894
819		- if (field_avail(typeof(resp), rss_caps, uhw->outlen)) {
	895	+ if (offsetofend(typeof(resp), rss_caps) <= uhw_outlen) {
820	896	resp.rss_caps.rx_hash_function =
821	897	MLX5_RX_HASH_FUNC_TOEPLITZ;
822	898	resp.rss_caps.rx_hash_fields_mask =
..	..	@@ -836,9 +912,9 @@
836	912	resp.response_length += sizeof(resp.rss_caps);
837	913	}
838	914	} else {
839		- if (field_avail(typeof(resp), tso_caps, uhw->outlen))
	915	+ if (offsetofend(typeof(resp), tso_caps) <= uhw_outlen)
840	916	resp.response_length += sizeof(resp.tso_caps);
841		- if (field_avail(typeof(resp), rss_caps, uhw->outlen))
	917	+ if (offsetofend(typeof(resp), rss_caps) <= uhw_outlen)
842	918	resp.response_length += sizeof(resp.rss_caps);
843	919	}
844	920
..	..	@@ -904,27 +980,47 @@
904	980	props->max_srq_sge = max_rq_sg - 1;
905	981	props->max_fast_reg_page_list_len =
906	982	1 << MLX5_CAP_GEN(mdev, log_max_klm_list_size);
	983	+ props->max_pi_fast_reg_page_list_len =
	984	+ props->max_fast_reg_page_list_len / 2;
	985	+ props->max_sgl_rd =
	986	+ MLX5_CAP_GEN(mdev, max_sgl_for_optimized_performance);
907	987	get_atomic_caps_qp(dev, props);
908	988	props->masked_atomic_cap = IB_ATOMIC_NONE;
909	989	props->max_mcast_grp = 1 << MLX5_CAP_GEN(mdev, log_max_mcg);
910	990	props->max_mcast_qp_attach = MLX5_CAP_GEN(mdev, max_qp_mcg);
911	991	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
912	992	props->max_mcast_grp;
913		- props->max_map_per_fmr = INT_MAX; /* no limit in ConnectIB */
914	993	props->max_ah = INT_MAX;
915	994	props->hca_core_clock = MLX5_CAP_GEN(mdev, device_frequency_khz);
916	995	props->timestamp_mask = 0x7FFFFFFFFFFFFFFFULL;
917	996
918		-#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
919		- if (MLX5_CAP_GEN(mdev, pg))
920		- props->device_cap_flags \|= IB_DEVICE_ON_DEMAND_PAGING;
921		- props->odp_caps = dev->odp_caps;
922		-#endif
	997	+ if (IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING)) {
	998	+ if (dev->odp_caps.general_caps & IB_ODP_SUPPORT)
	999	+ props->device_cap_flags \|= IB_DEVICE_ON_DEMAND_PAGING;
	1000	+ props->odp_caps = dev->odp_caps;
	1001	+ if (!uhw) {
	1002	+ /* ODP for kernel QPs is not implemented for receive
	1003	+ * WQEs and SRQ WQEs
	1004	+ */
	1005	+ props->odp_caps.per_transport_caps.rc_odp_caps &=
	1006	+ ~(IB_ODP_SUPPORT_READ \|
	1007	+ IB_ODP_SUPPORT_SRQ_RECV);
	1008	+ props->odp_caps.per_transport_caps.uc_odp_caps &=
	1009	+ ~(IB_ODP_SUPPORT_READ \|
	1010	+ IB_ODP_SUPPORT_SRQ_RECV);
	1011	+ props->odp_caps.per_transport_caps.ud_odp_caps &=
	1012	+ ~(IB_ODP_SUPPORT_READ \|
	1013	+ IB_ODP_SUPPORT_SRQ_RECV);
	1014	+ props->odp_caps.per_transport_caps.xrc_odp_caps &=
	1015	+ ~(IB_ODP_SUPPORT_READ \|
	1016	+ IB_ODP_SUPPORT_SRQ_RECV);
	1017	+ }
	1018	+ }
923	1019
924	1020	if (MLX5_CAP_GEN(mdev, cd))
925	1021	props->device_cap_flags \|= IB_DEVICE_CROSS_CHANNEL;
926	1022
927		- if (!mlx5_core_is_pf(mdev))
	1023	+ if (mlx5_core_is_vf(mdev))
928	1024	props->device_cap_flags \|= IB_DEVICE_VIRTUAL_FUNCTION;
929	1025
930	1026	if (mlx5_ib_port_link_layer(ibdev, 1) ==
..	..	@@ -959,7 +1055,7 @@
959	1055	MLX5_MAX_CQ_PERIOD;
960	1056	}
961	1057
962		- if (field_avail(typeof(resp), cqe_comp_caps, uhw->outlen)) {
	1058	+ if (offsetofend(typeof(resp), cqe_comp_caps) <= uhw_outlen) {
963	1059	resp.response_length += sizeof(resp.cqe_comp_caps);
964	1060
965	1061	if (MLX5_CAP_GEN(dev->mdev, cqe_compression)) {
..	..	@@ -977,7 +1073,7 @@
977	1073	}
978	1074	}
979	1075
980		- if (field_avail(typeof(resp), packet_pacing_caps, uhw->outlen) &&
	1076	+ if (offsetofend(typeof(resp), packet_pacing_caps) <= uhw_outlen &&
981	1077	raw_support) {
982	1078	if (MLX5_CAP_QOS(mdev, packet_pacing) &&
983	1079	MLX5_CAP_GEN(mdev, qos)) {
..	..	@@ -995,8 +1091,8 @@
995	1091	resp.response_length += sizeof(resp.packet_pacing_caps);
996	1092	}
997	1093
998		- if (field_avail(typeof(resp), mlx5_ib_support_multi_pkt_send_wqes,
999		- uhw->outlen)) {
	1094	+ if (offsetofend(typeof(resp), mlx5_ib_support_multi_pkt_send_wqes) <=
	1095	+ uhw_outlen) {
1000	1096	if (MLX5_CAP_ETH(mdev, multi_pkt_send_wqe))
1001	1097	resp.mlx5_ib_support_multi_pkt_send_wqes =
1002	1098	MLX5_IB_ALLOW_MPW;
..	..	@@ -1009,7 +1105,7 @@
1009	1105	sizeof(resp.mlx5_ib_support_multi_pkt_send_wqes);
1010	1106	}
1011	1107
1012		- if (field_avail(typeof(resp), flags, uhw->outlen)) {
	1108	+ if (offsetofend(typeof(resp), flags) <= uhw_outlen) {
1013	1109	resp.response_length += sizeof(resp.flags);
1014	1110
1015	1111	if (MLX5_CAP_GEN(mdev, cqe_compression_128))
..	..	@@ -1018,10 +1114,14 @@
1018	1114
1019	1115	if (MLX5_CAP_GEN(mdev, cqe_128_always))
1020	1116	resp.flags \|= MLX5_IB_QUERY_DEV_RESP_FLAGS_CQE_128B_PAD;
	1117	+ if (MLX5_CAP_GEN(mdev, qp_packet_based))
	1118	+ resp.flags \|=
	1119	+ MLX5_IB_QUERY_DEV_RESP_PACKET_BASED_CREDIT_MODE;
	1120	+
	1121	+ resp.flags \|= MLX5_IB_QUERY_DEV_RESP_FLAGS_SCAT2CQE_DCT;
1021	1122	}
1022	1123
1023		- if (field_avail(typeof(resp), sw_parsing_caps,
1024		- uhw->outlen)) {
	1124	+ if (offsetofend(typeof(resp), sw_parsing_caps) <= uhw_outlen) {
1025	1125	resp.response_length += sizeof(resp.sw_parsing_caps);
1026	1126	if (MLX5_CAP_ETH(mdev, swp)) {
1027	1127	resp.sw_parsing_caps.sw_parsing_offloads \|=
..	..	@@ -1041,7 +1141,7 @@
1041	1141	}
1042	1142	}
1043	1143
1044		- if (field_avail(typeof(resp), striding_rq_caps, uhw->outlen) &&
	1144	+ if (offsetofend(typeof(resp), striding_rq_caps) <= uhw_outlen &&
1045	1145	raw_support) {
1046	1146	resp.response_length += sizeof(resp.striding_rq_caps);
1047	1147	if (MLX5_CAP_GEN(mdev, striding_rq)) {
..	..	@@ -1049,8 +1149,14 @@
1049	1149	MLX5_MIN_SINGLE_STRIDE_LOG_NUM_BYTES;
1050	1150	resp.striding_rq_caps.max_single_stride_log_num_of_bytes =
1051	1151	MLX5_MAX_SINGLE_STRIDE_LOG_NUM_BYTES;
1052		- resp.striding_rq_caps.min_single_wqe_log_num_of_strides =
1053		- MLX5_MIN_SINGLE_WQE_LOG_NUM_STRIDES;
	1152	+ if (MLX5_CAP_GEN(dev->mdev, ext_stride_num_range))
	1153	+ resp.striding_rq_caps
	1154	+ .min_single_wqe_log_num_of_strides =
	1155	+ MLX5_EXT_MIN_SINGLE_WQE_LOG_NUM_STRIDES;
	1156	+ else
	1157	+ resp.striding_rq_caps
	1158	+ .min_single_wqe_log_num_of_strides =
	1159	+ MLX5_MIN_SINGLE_WQE_LOG_NUM_STRIDES;
1054	1160	resp.striding_rq_caps.max_single_wqe_log_num_of_strides =
1055	1161	MLX5_MAX_SINGLE_WQE_LOG_NUM_STRIDES;
1056	1162	resp.striding_rq_caps.supported_qpts =
..	..	@@ -1058,8 +1164,7 @@
1058	1164	}
1059	1165	}
1060	1166
1061		- if (field_avail(typeof(resp), tunnel_offloads_caps,
1062		- uhw->outlen)) {
	1167	+ if (offsetofend(typeof(resp), tunnel_offloads_caps) <= uhw_outlen) {
1063	1168	resp.response_length += sizeof(resp.tunnel_offloads_caps);
1064	1169	if (MLX5_CAP_ETH(mdev, tunnel_stateless_vxlan))
1065	1170	resp.tunnel_offloads_caps \|=
..	..	@@ -1078,7 +1183,7 @@
1078	1183	MLX5_IB_TUNNELED_OFFLOADS_MPLS_UDP;
1079	1184	}
1080	1185
1081		- if (uhw->outlen) {
	1186	+ if (uhw_outlen) {
1082	1187	err = ib_copy_to_udata(uhw, &resp, resp.response_length);
1083	1188
1084	1189	if (err)
..	..	@@ -1088,30 +1193,24 @@
1088	1193	return 0;
1089	1194	}
1090	1195
1091		-enum mlx5_ib_width {
1092		- MLX5_IB_WIDTH_1X = 1 << 0,
1093		- MLX5_IB_WIDTH_2X = 1 << 1,
1094		- MLX5_IB_WIDTH_4X = 1 << 2,
1095		- MLX5_IB_WIDTH_8X = 1 << 3,
1096		- MLX5_IB_WIDTH_12X = 1 << 4
1097		-};
1098		-
1099		-static void translate_active_width(struct ib_device *ibdev, u8 active_width,
1100		- u8 *ib_width)
	1196	+static void translate_active_width(struct ib_device *ibdev, u16 active_width,
	1197	+ u8 *ib_width)
1101	1198	{
1102	1199	struct mlx5_ib_dev *dev = to_mdev(ibdev);
1103	1200
1104		- if (active_width & MLX5_IB_WIDTH_1X)
	1201	+ if (active_width & MLX5_PTYS_WIDTH_1X)
1105	1202	*ib_width = IB_WIDTH_1X;
1106		- else if (active_width & MLX5_IB_WIDTH_4X)
	1203	+ else if (active_width & MLX5_PTYS_WIDTH_2X)
	1204	+ *ib_width = IB_WIDTH_2X;
	1205	+ else if (active_width & MLX5_PTYS_WIDTH_4X)
1107	1206	*ib_width = IB_WIDTH_4X;
1108		- else if (active_width & MLX5_IB_WIDTH_8X)
	1207	+ else if (active_width & MLX5_PTYS_WIDTH_8X)
1109	1208	*ib_width = IB_WIDTH_8X;
1110		- else if (active_width & MLX5_IB_WIDTH_12X)
	1209	+ else if (active_width & MLX5_PTYS_WIDTH_12X)
1111	1210	*ib_width = IB_WIDTH_12X;
1112	1211	else {
1113	1212	mlx5_ib_dbg(dev, "Invalid active_width %d, setting width to default value: 4x\n",
1114		- (int)active_width);
	1213	+ active_width);
1115	1214	*ib_width = IB_WIDTH_4X;
1116	1215	}
1117	1216
..	..	@@ -1188,7 +1287,7 @@
1188	1287	u16 max_mtu;
1189	1288	u16 oper_mtu;
1190	1289	int err;
1191		- u8 ib_link_width_oper;
	1290	+ u16 ib_link_width_oper;
1192	1291	u8 vl_hw_cap;
1193	1292
1194	1293	rep = kzalloc(sizeof(*rep), GFP_KERNEL);
..	..	@@ -1218,15 +1317,15 @@
1218	1317	props->subnet_timeout = rep->subnet_timeout;
1219	1318	props->init_type_reply = rep->init_type_reply;
1220	1319
1221		- err = mlx5_query_port_link_width_oper(mdev, &ib_link_width_oper, port);
	1320	+ if (props->port_cap_flags & IB_PORT_CAP_MASK2_SUP)
	1321	+ props->port_cap_flags2 = rep->cap_mask2;
	1322	+
	1323	+ err = mlx5_query_ib_port_oper(mdev, &ib_link_width_oper,
	1324	+ &props->active_speed, port);
1222	1325	if (err)
1223	1326	goto out;
1224	1327
1225	1328	translate_active_width(ibdev, ib_link_width_oper, &props->active_width);
1226		-
1227		- err = mlx5_query_port_ib_proto_oper(mdev, &props->active_speed, port);
1228		- if (err)
1229		- goto out;
1230	1329
1231	1330	mlx5_query_port_max_mtu(mdev, &max_mtu, port);
1232	1331
..	..	@@ -1297,7 +1396,9 @@
1297	1396	{
1298	1397	int ret;
1299	1398
1300		- /* Only link layer == ethernet is valid for representors */
	1399	+ /* Only link layer == ethernet is valid for representors
	1400	+ * and we always use port 1
	1401	+ */
1301	1402	ret = mlx5_query_port_roce(ibdev, port, props);
1302	1403	if (ret \|\| !props)
1303	1404	return ret;
..	..	@@ -1566,14 +1667,57 @@
1566	1667	mlx5_cmd_free_uar(dev->mdev, bfregi->sys_pages[i]);
1567	1668	}
1568	1669
1569		-static int mlx5_ib_alloc_transport_domain(struct mlx5_ib_dev dev, u32 tdn)
	1670	+int mlx5_ib_enable_lb(struct mlx5_ib_dev *dev, bool td, bool qp)
	1671	+{
	1672	+ int err = 0;
	1673	+
	1674	+ mutex_lock(&dev->lb.mutex);
	1675	+ if (td)
	1676	+ dev->lb.user_td++;
	1677	+ if (qp)
	1678	+ dev->lb.qps++;
	1679	+
	1680	+ if (dev->lb.user_td == 2 \|\|
	1681	+ dev->lb.qps == 1) {
	1682	+ if (!dev->lb.enabled) {
	1683	+ err = mlx5_nic_vport_update_local_lb(dev->mdev, true);
	1684	+ dev->lb.enabled = true;
	1685	+ }
	1686	+ }
	1687	+
	1688	+ mutex_unlock(&dev->lb.mutex);
	1689	+
	1690	+ return err;
	1691	+}
	1692	+
	1693	+void mlx5_ib_disable_lb(struct mlx5_ib_dev *dev, bool td, bool qp)
	1694	+{
	1695	+ mutex_lock(&dev->lb.mutex);
	1696	+ if (td)
	1697	+ dev->lb.user_td--;
	1698	+ if (qp)
	1699	+ dev->lb.qps--;
	1700	+
	1701	+ if (dev->lb.user_td == 1 &&
	1702	+ dev->lb.qps == 0) {
	1703	+ if (dev->lb.enabled) {
	1704	+ mlx5_nic_vport_update_local_lb(dev->mdev, false);
	1705	+ dev->lb.enabled = false;
	1706	+ }
	1707	+ }
	1708	+
	1709	+ mutex_unlock(&dev->lb.mutex);
	1710	+}
	1711	+
	1712	+static int mlx5_ib_alloc_transport_domain(struct mlx5_ib_dev dev, u32 tdn,
	1713	+ u16 uid)
1570	1714	{
1571	1715	int err;
1572	1716
1573	1717	if (!MLX5_CAP_GEN(dev->mdev, log_max_transport_domain))
1574	1718	return 0;
1575	1719
1576		- err = mlx5_core_alloc_transport_domain(dev->mdev, tdn);
	1720	+ err = mlx5_cmd_alloc_transport_domain(dev->mdev, tdn, uid);
1577	1721	if (err)
1578	1722	return err;
1579	1723
..	..	@@ -1582,115 +1726,160 @@
1582	1726	!MLX5_CAP_GEN(dev->mdev, disable_local_lb_mc)))
1583	1727	return err;
1584	1728
1585		- mutex_lock(&dev->lb_mutex);
1586		- dev->user_td++;
1587		-
1588		- if (dev->user_td == 2)
1589		- err = mlx5_nic_vport_update_local_lb(dev->mdev, true);
1590		-
1591		- mutex_unlock(&dev->lb_mutex);
1592		- return err;
	1729	+ return mlx5_ib_enable_lb(dev, true, false);
1593	1730	}
1594	1731
1595		-static void mlx5_ib_dealloc_transport_domain(struct mlx5_ib_dev *dev, u32 tdn)
	1732	+static void mlx5_ib_dealloc_transport_domain(struct mlx5_ib_dev *dev, u32 tdn,
	1733	+ u16 uid)
1596	1734	{
1597	1735	if (!MLX5_CAP_GEN(dev->mdev, log_max_transport_domain))
1598	1736	return;
1599	1737
1600		- mlx5_core_dealloc_transport_domain(dev->mdev, tdn);
	1738	+ mlx5_cmd_dealloc_transport_domain(dev->mdev, tdn, uid);
1601	1739
1602	1740	if ((MLX5_CAP_GEN(dev->mdev, port_type) != MLX5_CAP_PORT_TYPE_ETH) \|\|
1603	1741	(!MLX5_CAP_GEN(dev->mdev, disable_local_lb_uc) &&
1604	1742	!MLX5_CAP_GEN(dev->mdev, disable_local_lb_mc)))
1605	1743	return;
1606	1744
1607		- mutex_lock(&dev->lb_mutex);
1608		- dev->user_td--;
1609		-
1610		- if (dev->user_td < 2)
1611		- mlx5_nic_vport_update_local_lb(dev->mdev, false);
1612		-
1613		- mutex_unlock(&dev->lb_mutex);
	1745	+ mlx5_ib_disable_lb(dev, true, false);
1614	1746	}
1615	1747
1616		-static struct ib_ucontext mlx5_ib_alloc_ucontext(struct ib_device ibdev,
1617		- struct ib_udata *udata)
	1748	+static int set_ucontext_resp(struct ib_ucontext *uctx,
	1749	+ struct mlx5_ib_alloc_ucontext_resp *resp)
1618	1750	{
	1751	+ struct ib_device *ibdev = uctx->device;
	1752	+ struct mlx5_ib_dev *dev = to_mdev(ibdev);
	1753	+ struct mlx5_ib_ucontext *context = to_mucontext(uctx);
	1754	+ struct mlx5_bfreg_info *bfregi = &context->bfregi;
	1755	+ int err;
	1756	+
	1757	+ if (MLX5_CAP_GEN(dev->mdev, dump_fill_mkey)) {
	1758	+ err = mlx5_cmd_dump_fill_mkey(dev->mdev,
	1759	+ &resp->dump_fill_mkey);
	1760	+ if (err)
	1761	+ return err;
	1762	+ resp->comp_mask \|=
	1763	+ MLX5_IB_ALLOC_UCONTEXT_RESP_MASK_DUMP_FILL_MKEY;
	1764	+ }
	1765	+
	1766	+ resp->qp_tab_size = 1 << MLX5_CAP_GEN(dev->mdev, log_max_qp);
	1767	+ if (dev->wc_support)
	1768	+ resp->bf_reg_size = 1 << MLX5_CAP_GEN(dev->mdev,
	1769	+ log_bf_reg_size);
	1770	+ resp->cache_line_size = cache_line_size();
	1771	+ resp->max_sq_desc_sz = MLX5_CAP_GEN(dev->mdev, max_wqe_sz_sq);
	1772	+ resp->max_rq_desc_sz = MLX5_CAP_GEN(dev->mdev, max_wqe_sz_rq);
	1773	+ resp->max_send_wqebb = 1 << MLX5_CAP_GEN(dev->mdev, log_max_qp_sz);
	1774	+ resp->max_recv_wr = 1 << MLX5_CAP_GEN(dev->mdev, log_max_qp_sz);
	1775	+ resp->max_srq_recv_wr = 1 << MLX5_CAP_GEN(dev->mdev, log_max_srq_sz);
	1776	+ resp->cqe_version = context->cqe_version;
	1777	+ resp->log_uar_size = MLX5_CAP_GEN(dev->mdev, uar_4k) ?
	1778	+ MLX5_ADAPTER_PAGE_SHIFT : PAGE_SHIFT;
	1779	+ resp->num_uars_per_page = MLX5_CAP_GEN(dev->mdev, uar_4k) ?
	1780	+ MLX5_CAP_GEN(dev->mdev,
	1781	+ num_of_uars_per_page) : 1;
	1782	+
	1783	+ if (mlx5_accel_ipsec_device_caps(dev->mdev) &
	1784	+ MLX5_ACCEL_IPSEC_CAP_DEVICE) {
	1785	+ if (mlx5_get_flow_namespace(dev->mdev,
	1786	+ MLX5_FLOW_NAMESPACE_EGRESS))
	1787	+ resp->flow_action_flags \|= MLX5_USER_ALLOC_UCONTEXT_FLOW_ACTION_FLAGS_ESP_AES_GCM;
	1788	+ if (mlx5_accel_ipsec_device_caps(dev->mdev) &
	1789	+ MLX5_ACCEL_IPSEC_CAP_REQUIRED_METADATA)
	1790	+ resp->flow_action_flags \|= MLX5_USER_ALLOC_UCONTEXT_FLOW_ACTION_FLAGS_ESP_AES_GCM_REQ_METADATA;
	1791	+ if (MLX5_CAP_FLOWTABLE(dev->mdev, flow_table_properties_nic_receive.ft_field_support.outer_esp_spi))
	1792	+ resp->flow_action_flags \|= MLX5_USER_ALLOC_UCONTEXT_FLOW_ACTION_FLAGS_ESP_AES_GCM_SPI_STEERING;
	1793	+ if (mlx5_accel_ipsec_device_caps(dev->mdev) &
	1794	+ MLX5_ACCEL_IPSEC_CAP_TX_IV_IS_ESN)
	1795	+ resp->flow_action_flags \|= MLX5_USER_ALLOC_UCONTEXT_FLOW_ACTION_FLAGS_ESP_AES_GCM_TX_IV_IS_ESN;
	1796	+ /* MLX5_USER_ALLOC_UCONTEXT_FLOW_ACTION_FLAGS_ESP_AES_GCM_FULL_OFFLOAD is currently always 0 */
	1797	+ }
	1798	+
	1799	+ resp->tot_bfregs = bfregi->lib_uar_dyn ? 0 :
	1800	+ bfregi->total_num_bfregs - bfregi->num_dyn_bfregs;
	1801	+ resp->num_ports = dev->num_ports;
	1802	+ resp->cmds_supp_uhw \|= MLX5_USER_CMDS_SUPP_UHW_QUERY_DEVICE \|
	1803	+ MLX5_USER_CMDS_SUPP_UHW_CREATE_AH;
	1804	+
	1805	+ if (mlx5_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET) {
	1806	+ mlx5_query_min_inline(dev->mdev, &resp->eth_min_inline);
	1807	+ resp->eth_min_inline++;
	1808	+ }
	1809	+
	1810	+ if (dev->mdev->clock_info)
	1811	+ resp->clock_info_versions = BIT(MLX5_IB_CLOCK_INFO_V1);
	1812	+
	1813	+ /*
	1814	+ * We don't want to expose information from the PCI bar that is located
	1815	+ * after 4096 bytes, so if the arch only supports larger pages, let's
	1816	+ * pretend we don't support reading the HCA's core clock. This is also
	1817	+ * forced by mmap function.
	1818	+ */
	1819	+ if (PAGE_SIZE <= 4096) {
	1820	+ resp->comp_mask \|=
	1821	+ MLX5_IB_ALLOC_UCONTEXT_RESP_MASK_CORE_CLOCK_OFFSET;
	1822	+ resp->hca_core_clock_offset =
	1823	+ offsetof(struct mlx5_init_seg,
	1824	+ internal_timer_h) % PAGE_SIZE;
	1825	+ }
	1826	+
	1827	+ if (MLX5_CAP_GEN(dev->mdev, ece_support))
	1828	+ resp->comp_mask \|= MLX5_IB_ALLOC_UCONTEXT_RESP_MASK_ECE;
	1829	+
	1830	+ resp->num_dyn_bfregs = bfregi->num_dyn_bfregs;
	1831	+ return 0;
	1832	+}
	1833	+
	1834	+static int mlx5_ib_alloc_ucontext(struct ib_ucontext *uctx,
	1835	+ struct ib_udata *udata)
	1836	+{
	1837	+ struct ib_device *ibdev = uctx->device;
1619	1838	struct mlx5_ib_dev *dev = to_mdev(ibdev);
1620	1839	struct mlx5_ib_alloc_ucontext_req_v2 req = {};
1621	1840	struct mlx5_ib_alloc_ucontext_resp resp = {};
1622		- struct mlx5_core_dev *mdev = dev->mdev;
1623		- struct mlx5_ib_ucontext *context;
	1841	+ struct mlx5_ib_ucontext *context = to_mucontext(uctx);
1624	1842	struct mlx5_bfreg_info *bfregi;
1625	1843	int ver;
1626	1844	int err;
1627	1845	size_t min_req_v2 = offsetof(struct mlx5_ib_alloc_ucontext_req_v2,
1628	1846	max_cqe_version);
1629		- u32 dump_fill_mkey;
1630	1847	bool lib_uar_4k;
	1848	+ bool lib_uar_dyn;
1631	1849
1632	1850	if (!dev->ib_active)
1633		- return ERR_PTR(-EAGAIN);
	1851	+ return -EAGAIN;
1634	1852
1635	1853	if (udata->inlen == sizeof(struct mlx5_ib_alloc_ucontext_req))
1636	1854	ver = 0;
1637	1855	else if (udata->inlen >= min_req_v2)
1638	1856	ver = 2;
1639	1857	else
1640		- return ERR_PTR(-EINVAL);
	1858	+ return -EINVAL;
1641	1859
1642	1860	err = ib_copy_from_udata(&req, udata, min(udata->inlen, sizeof(req)));
1643	1861	if (err)
1644		- return ERR_PTR(err);
	1862	+ return err;
1645	1863
1646	1864	if (req.flags & ~MLX5_IB_ALLOC_UCTX_DEVX)
1647		- return ERR_PTR(-EOPNOTSUPP);
	1865	+ return -EOPNOTSUPP;
1648	1866
1649	1867	if (req.comp_mask \|\| req.reserved0 \|\| req.reserved1 \|\| req.reserved2)
1650		- return ERR_PTR(-EOPNOTSUPP);
	1868	+ return -EOPNOTSUPP;
1651	1869
1652	1870	req.total_num_bfregs = ALIGN(req.total_num_bfregs,
1653	1871	MLX5_NON_FP_BFREGS_PER_UAR);
1654	1872	if (req.num_low_latency_bfregs > req.total_num_bfregs - 1)
1655		- return ERR_PTR(-EINVAL);
1656		-
1657		- resp.qp_tab_size = 1 << MLX5_CAP_GEN(dev->mdev, log_max_qp);
1658		- if (mlx5_core_is_pf(dev->mdev) && MLX5_CAP_GEN(dev->mdev, bf))
1659		- resp.bf_reg_size = 1 << MLX5_CAP_GEN(dev->mdev, log_bf_reg_size);
1660		- resp.cache_line_size = cache_line_size();
1661		- resp.max_sq_desc_sz = MLX5_CAP_GEN(dev->mdev, max_wqe_sz_sq);
1662		- resp.max_rq_desc_sz = MLX5_CAP_GEN(dev->mdev, max_wqe_sz_rq);
1663		- resp.max_send_wqebb = 1 << MLX5_CAP_GEN(dev->mdev, log_max_qp_sz);
1664		- resp.max_recv_wr = 1 << MLX5_CAP_GEN(dev->mdev, log_max_qp_sz);
1665		- resp.max_srq_recv_wr = 1 << MLX5_CAP_GEN(dev->mdev, log_max_srq_sz);
1666		- resp.cqe_version = min_t(__u8,
1667		- (__u8)MLX5_CAP_GEN(dev->mdev, cqe_version),
1668		- req.max_cqe_version);
1669		- resp.log_uar_size = MLX5_CAP_GEN(dev->mdev, uar_4k) ?
1670		- MLX5_ADAPTER_PAGE_SHIFT : PAGE_SHIFT;
1671		- resp.num_uars_per_page = MLX5_CAP_GEN(dev->mdev, uar_4k) ?
1672		- MLX5_CAP_GEN(dev->mdev, num_of_uars_per_page) : 1;
1673		- resp.response_length = min(offsetof(typeof(resp), response_length) +
1674		- sizeof(resp.response_length), udata->outlen);
1675		-
1676		- if (mlx5_accel_ipsec_device_caps(dev->mdev) & MLX5_ACCEL_IPSEC_CAP_DEVICE) {
1677		- if (mlx5_get_flow_namespace(dev->mdev, MLX5_FLOW_NAMESPACE_EGRESS))
1678		- resp.flow_action_flags \|= MLX5_USER_ALLOC_UCONTEXT_FLOW_ACTION_FLAGS_ESP_AES_GCM;
1679		- if (mlx5_accel_ipsec_device_caps(dev->mdev) & MLX5_ACCEL_IPSEC_CAP_REQUIRED_METADATA)
1680		- resp.flow_action_flags \|= MLX5_USER_ALLOC_UCONTEXT_FLOW_ACTION_FLAGS_ESP_AES_GCM_REQ_METADATA;
1681		- if (MLX5_CAP_FLOWTABLE(dev->mdev, flow_table_properties_nic_receive.ft_field_support.outer_esp_spi))
1682		- resp.flow_action_flags \|= MLX5_USER_ALLOC_UCONTEXT_FLOW_ACTION_FLAGS_ESP_AES_GCM_SPI_STEERING;
1683		- if (mlx5_accel_ipsec_device_caps(dev->mdev) & MLX5_ACCEL_IPSEC_CAP_TX_IV_IS_ESN)
1684		- resp.flow_action_flags \|= MLX5_USER_ALLOC_UCONTEXT_FLOW_ACTION_FLAGS_ESP_AES_GCM_TX_IV_IS_ESN;
1685		- /* MLX5_USER_ALLOC_UCONTEXT_FLOW_ACTION_FLAGS_ESP_AES_GCM_FULL_OFFLOAD is currently always 0 */
1686		- }
1687		-
1688		- context = kzalloc(sizeof(*context), GFP_KERNEL);
1689		- if (!context)
1690		- return ERR_PTR(-ENOMEM);
	1873	+ return -EINVAL;
1691	1874
1692	1875	lib_uar_4k = req.lib_caps & MLX5_LIB_CAP_4K_UAR;
	1876	+ lib_uar_dyn = req.lib_caps & MLX5_LIB_CAP_DYN_UAR;
1693	1877	bfregi = &context->bfregi;
	1878	+
	1879	+ if (lib_uar_dyn) {
	1880	+ bfregi->lib_uar_dyn = lib_uar_dyn;
	1881	+ goto uar_done;
	1882	+ }
1694	1883
1695	1884	/* updates req->total_num_bfregs */
1696	1885	err = calc_total_bfregs(dev, lib_uar_4k, &req, bfregi);
..	..	@@ -1718,124 +1907,55 @@
1718	1907	if (err)
1719	1908	goto out_sys_pages;
1720	1909
1721		-#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1722		- context->ibucontext.invalidate_range = &mlx5_ib_invalidate_range;
1723		-#endif
1724		-
1725		- err = mlx5_ib_alloc_transport_domain(dev, &context->tdn);
1726		- if (err)
1727		- goto out_uars;
1728		-
	1910	+uar_done:
1729	1911	if (req.flags & MLX5_IB_ALLOC_UCTX_DEVX) {
1730		- /* Block DEVX on Infiniband as of SELinux */
1731		- if (mlx5_ib_port_link_layer(ibdev, 1) != IB_LINK_LAYER_ETHERNET) {
1732		- err = -EPERM;
1733		- goto out_td;
1734		- }
1735		-
1736		- err = mlx5_ib_devx_create(dev, context);
1737		- if (err)
1738		- goto out_td;
	1912	+ err = mlx5_ib_devx_create(dev, true);
	1913	+ if (err < 0)
	1914	+ goto out_uars;
	1915	+ context->devx_uid = err;
1739	1916	}
1740	1917
1741		- if (MLX5_CAP_GEN(dev->mdev, dump_fill_mkey)) {
1742		- err = mlx5_cmd_dump_fill_mkey(dev->mdev, &dump_fill_mkey);
1743		- if (err)
1744		- goto out_mdev;
1745		- }
	1918	+ err = mlx5_ib_alloc_transport_domain(dev, &context->tdn,
	1919	+ context->devx_uid);
	1920	+ if (err)
	1921	+ goto out_devx;
1746	1922
1747		- INIT_LIST_HEAD(&context->vma_private_list);
1748		- mutex_init(&context->vma_private_list_mutex);
1749	1923	INIT_LIST_HEAD(&context->db_page_list);
1750	1924	mutex_init(&context->db_page_mutex);
1751	1925
1752		- resp.tot_bfregs = req.total_num_bfregs;
1753		- resp.num_ports = dev->num_ports;
	1926	+ context->cqe_version = min_t(__u8,
	1927	+ (__u8)MLX5_CAP_GEN(dev->mdev, cqe_version),
	1928	+ req.max_cqe_version);
1754	1929
1755		- if (field_avail(typeof(resp), cqe_version, udata->outlen))
1756		- resp.response_length += sizeof(resp.cqe_version);
	1930	+ err = set_ucontext_resp(uctx, &resp);
	1931	+ if (err)
	1932	+ goto out_mdev;
1757	1933
1758		- if (field_avail(typeof(resp), cmds_supp_uhw, udata->outlen)) {
1759		- resp.cmds_supp_uhw \|= MLX5_USER_CMDS_SUPP_UHW_QUERY_DEVICE \|
1760		- MLX5_USER_CMDS_SUPP_UHW_CREATE_AH;
1761		- resp.response_length += sizeof(resp.cmds_supp_uhw);
1762		- }
1763		-
1764		- if (field_avail(typeof(resp), eth_min_inline, udata->outlen)) {
1765		- if (mlx5_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET) {
1766		- mlx5_query_min_inline(dev->mdev, &resp.eth_min_inline);
1767		- resp.eth_min_inline++;
1768		- }
1769		- resp.response_length += sizeof(resp.eth_min_inline);
1770		- }
1771		-
1772		- if (field_avail(typeof(resp), clock_info_versions, udata->outlen)) {
1773		- if (mdev->clock_info)
1774		- resp.clock_info_versions = BIT(MLX5_IB_CLOCK_INFO_V1);
1775		- resp.response_length += sizeof(resp.clock_info_versions);
1776		- }
1777		-
1778		- /*
1779		- * We don't want to expose information from the PCI bar that is located
1780		- * after 4096 bytes, so if the arch only supports larger pages, let's
1781		- * pretend we don't support reading the HCA's core clock. This is also
1782		- * forced by mmap function.
1783		- */
1784		- if (field_avail(typeof(resp), hca_core_clock_offset, udata->outlen)) {
1785		- if (PAGE_SIZE <= 4096) {
1786		- resp.comp_mask \|=
1787		- MLX5_IB_ALLOC_UCONTEXT_RESP_MASK_CORE_CLOCK_OFFSET;
1788		- resp.hca_core_clock_offset =
1789		- offsetof(struct mlx5_init_seg, internal_timer_h) % PAGE_SIZE;
1790		- }
1791		- resp.response_length += sizeof(resp.hca_core_clock_offset);
1792		- }
1793		-
1794		- if (field_avail(typeof(resp), log_uar_size, udata->outlen))
1795		- resp.response_length += sizeof(resp.log_uar_size);
1796		-
1797		- if (field_avail(typeof(resp), num_uars_per_page, udata->outlen))
1798		- resp.response_length += sizeof(resp.num_uars_per_page);
1799		-
1800		- if (field_avail(typeof(resp), num_dyn_bfregs, udata->outlen)) {
1801		- resp.num_dyn_bfregs = bfregi->num_dyn_bfregs;
1802		- resp.response_length += sizeof(resp.num_dyn_bfregs);
1803		- }
1804		-
1805		- if (field_avail(typeof(resp), dump_fill_mkey, udata->outlen)) {
1806		- if (MLX5_CAP_GEN(dev->mdev, dump_fill_mkey)) {
1807		- resp.dump_fill_mkey = dump_fill_mkey;
1808		- resp.comp_mask \|=
1809		- MLX5_IB_ALLOC_UCONTEXT_RESP_MASK_DUMP_FILL_MKEY;
1810		- }
1811		- resp.response_length += sizeof(resp.dump_fill_mkey);
1812		- }
1813		-
	1934	+ resp.response_length = min(udata->outlen, sizeof(resp));
1814	1935	err = ib_copy_to_udata(udata, &resp, resp.response_length);
1815	1936	if (err)
1816	1937	goto out_mdev;
1817	1938
1818	1939	bfregi->ver = ver;
1819	1940	bfregi->num_low_latency_bfregs = req.num_low_latency_bfregs;
1820		- context->cqe_version = resp.cqe_version;
1821	1941	context->lib_caps = req.lib_caps;
1822	1942	print_lib_caps(dev, context->lib_caps);
1823	1943
1824		- if (mlx5_lag_is_active(dev->mdev)) {
1825		- u8 port = mlx5_core_native_port_num(dev->mdev);
	1944	+ if (mlx5_ib_lag_should_assign_affinity(dev)) {
	1945	+ u8 port = mlx5_core_native_port_num(dev->mdev) - 1;
1826	1946
1827	1947	atomic_set(&context->tx_port_affinity,
1828	1948	atomic_add_return(
1829		- 1, &dev->roce[port].tx_port_affinity));
	1949	+ 1, &dev->port[port].roce.tx_port_affinity));
1830	1950	}
1831	1951
1832		- return &context->ibucontext;
	1952	+ return 0;
1833	1953
1834	1954	out_mdev:
	1955	+ mlx5_ib_dealloc_transport_domain(dev, context->tdn, context->devx_uid);
	1956	+out_devx:
1835	1957	if (req.flags & MLX5_IB_ALLOC_UCTX_DEVX)
1836		- mlx5_ib_devx_destroy(dev, context);
1837		-out_td:
1838		- mlx5_ib_dealloc_transport_domain(dev, context->tdn);
	1958	+ mlx5_ib_devx_destroy(dev, context->devx_uid);
1839	1959
1840	1960	out_uars:
1841	1961	deallocate_uars(dev, context);
..	..	@@ -1847,29 +1967,47 @@
1847	1967	kfree(bfregi->count);
1848	1968
1849	1969	out_ctx:
1850		- kfree(context);
1851		-
1852		- return ERR_PTR(err);
	1970	+ return err;
1853	1971	}
1854	1972
1855		-static int mlx5_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
	1973	+static int mlx5_ib_query_ucontext(struct ib_ucontext *ibcontext,
	1974	+ struct uverbs_attr_bundle *attrs)
	1975	+{
	1976	+ struct mlx5_ib_alloc_ucontext_resp uctx_resp = {};
	1977	+ int ret;
	1978	+
	1979	+ ret = set_ucontext_resp(ibcontext, &uctx_resp);
	1980	+ if (ret)
	1981	+ return ret;
	1982	+
	1983	+ uctx_resp.response_length =
	1984	+ min_t(size_t,
	1985	+ uverbs_attr_get_len(attrs,
	1986	+ MLX5_IB_ATTR_QUERY_CONTEXT_RESP_UCTX),
	1987	+ sizeof(uctx_resp));
	1988	+
	1989	+ ret = uverbs_copy_to_struct_or_zero(attrs,
	1990	+ MLX5_IB_ATTR_QUERY_CONTEXT_RESP_UCTX,
	1991	+ &uctx_resp,
	1992	+ sizeof(uctx_resp));
	1993	+ return ret;
	1994	+}
	1995	+
	1996	+static void mlx5_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1856	1997	{
1857	1998	struct mlx5_ib_ucontext *context = to_mucontext(ibcontext);
1858	1999	struct mlx5_ib_dev *dev = to_mdev(ibcontext->device);
1859	2000	struct mlx5_bfreg_info *bfregi;
1860	2001
1861		- if (context->devx_uid)
1862		- mlx5_ib_devx_destroy(dev, context);
1863		-
1864	2002	bfregi = &context->bfregi;
1865		- mlx5_ib_dealloc_transport_domain(dev, context->tdn);
	2003	+ mlx5_ib_dealloc_transport_domain(dev, context->tdn, context->devx_uid);
	2004	+
	2005	+ if (context->devx_uid)
	2006	+ mlx5_ib_devx_destroy(dev, context->devx_uid);
1866	2007
1867	2008	deallocate_uars(dev, context);
1868	2009	kfree(bfregi->sys_pages);
1869	2010	kfree(bfregi->count);
1870		- kfree(context);
1871		-
1872		- return 0;
1873	2011	}
1874	2012
1875	2013	static phys_addr_t uar_index2pfn(struct mlx5_ib_dev *dev,
..	..	@@ -1879,7 +2017,18 @@
1879	2017
1880	2018	fw_uars_per_page = MLX5_CAP_GEN(dev->mdev, uar_4k) ? MLX5_UARS_IN_PAGE : 1;
1881	2019
1882		- return (pci_resource_start(dev->mdev->pdev, 0) >> PAGE_SHIFT) + uar_idx / fw_uars_per_page;
	2020	+ return (dev->mdev->bar_addr >> PAGE_SHIFT) + uar_idx / fw_uars_per_page;
	2021	+}
	2022	+
	2023	+static u64 uar_index2paddress(struct mlx5_ib_dev *dev,
	2024	+ int uar_idx)
	2025	+{
	2026	+ unsigned int fw_uars_per_page;
	2027	+
	2028	+ fw_uars_per_page = MLX5_CAP_GEN(dev->mdev, uar_4k) ?
	2029	+ MLX5_UARS_IN_PAGE : 1;
	2030	+
	2031	+ return (dev->mdev->bar_addr + (uar_idx / fw_uars_per_page) * PAGE_SIZE);
1883	2032	}
1884	2033
1885	2034	static int get_command(unsigned long offset)
..	..	@@ -1903,94 +2052,9 @@
1903	2052	return get_arg(offset) \| ((offset >> 16) & 0xff) << 8;
1904	2053	}
1905	2054
1906		-static void mlx5_ib_vma_open(struct vm_area_struct *area)
1907		-{
1908		- /* vma_open is called when a new VMA is created on top of our VMA. This
1909		- * is done through either mremap flow or split_vma (usually due to
1910		- * mlock, madvise, munmap, etc.) We do not support a clone of the VMA,
1911		- * as this VMA is strongly hardware related. Therefore we set the
1912		- * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
1913		- * calling us again and trying to do incorrect actions. We assume that
1914		- * the original VMA size is exactly a single page, and therefore all
1915		- * "splitting" operation will not happen to it.
1916		- */
1917		- area->vm_ops = NULL;
1918		-}
1919		-
1920		-static void mlx5_ib_vma_close(struct vm_area_struct *area)
1921		-{
1922		- struct mlx5_ib_vma_private_data *mlx5_ib_vma_priv_data;
1923		-
1924		- /* It's guaranteed that all VMAs opened on a FD are closed before the
1925		- * file itself is closed, therefore no sync is needed with the regular
1926		- * closing flow. (e.g. mlx5 ib_dealloc_ucontext)
1927		- * However need a sync with accessing the vma as part of
1928		- * mlx5_ib_disassociate_ucontext.
1929		- * The close operation is usually called under mm->mmap_sem except when
1930		- * process is exiting.
1931		- * The exiting case is handled explicitly as part of
1932		- * mlx5_ib_disassociate_ucontext.
1933		- */
1934		- mlx5_ib_vma_priv_data = (struct mlx5_ib_vma_private_data *)area->vm_private_data;
1935		-
1936		- /* setting the vma context pointer to null in the mlx5_ib driver's
1937		- * private data, to protect a race condition in
1938		- * mlx5_ib_disassociate_ucontext().
1939		- */
1940		- mlx5_ib_vma_priv_data->vma = NULL;
1941		- mutex_lock(mlx5_ib_vma_priv_data->vma_private_list_mutex);
1942		- list_del(&mlx5_ib_vma_priv_data->list);
1943		- mutex_unlock(mlx5_ib_vma_priv_data->vma_private_list_mutex);
1944		- kfree(mlx5_ib_vma_priv_data);
1945		-}
1946		-
1947		-static const struct vm_operations_struct mlx5_ib_vm_ops = {
1948		- .open = mlx5_ib_vma_open,
1949		- .close = mlx5_ib_vma_close
1950		-};
1951		-
1952		-static int mlx5_ib_set_vma_data(struct vm_area_struct *vma,
1953		- struct mlx5_ib_ucontext *ctx)
1954		-{
1955		- struct mlx5_ib_vma_private_data *vma_prv;
1956		- struct list_head *vma_head = &ctx->vma_private_list;
1957		-
1958		- vma_prv = kzalloc(sizeof(*vma_prv), GFP_KERNEL);
1959		- if (!vma_prv)
1960		- return -ENOMEM;
1961		-
1962		- vma_prv->vma = vma;
1963		- vma_prv->vma_private_list_mutex = &ctx->vma_private_list_mutex;
1964		- vma->vm_private_data = vma_prv;
1965		- vma->vm_ops = &mlx5_ib_vm_ops;
1966		-
1967		- mutex_lock(&ctx->vma_private_list_mutex);
1968		- list_add(&vma_prv->list, vma_head);
1969		- mutex_unlock(&ctx->vma_private_list_mutex);
1970		-
1971		- return 0;
1972		-}
1973	2055
1974	2056	static void mlx5_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1975	2057	{
1976		- struct vm_area_struct *vma;
1977		- struct mlx5_ib_vma_private_data vma_private, n;
1978		- struct mlx5_ib_ucontext *context = to_mucontext(ibcontext);
1979		-
1980		- mutex_lock(&context->vma_private_list_mutex);
1981		- list_for_each_entry_safe(vma_private, n, &context->vma_private_list,
1982		- list) {
1983		- vma = vma_private->vma;
1984		- zap_vma_ptes(vma, vma->vm_start, PAGE_SIZE);
1985		- /* context going to be destroyed, should
1986		- * not access ops any more.
1987		- */
1988		- vma->vm_flags &= ~(VM_SHARED \| VM_MAYSHARE);
1989		- vma->vm_ops = NULL;
1990		- list_del(&vma_private->list);
1991		- kfree(vma_private);
1992		- }
1993		- mutex_unlock(&context->vma_private_list_mutex);
1994	2058	}
1995	2059
1996	2060	static inline char *mmap_cmd2str(enum mlx5_ib_mmap_cmd cmd)
..	..	@@ -2005,7 +2069,7 @@
2005	2069	case MLX5_IB_MMAP_DEVICE_MEM:
2006	2070	return "Device Memory";
2007	2071	default:
2008		- return NULL;
	2072	+ return "Unknown";
2009	2073	}
2010	2074	}
2011	2075
..	..	@@ -2013,29 +2077,52 @@
2013	2077	struct vm_area_struct *vma,
2014	2078	struct mlx5_ib_ucontext *context)
2015	2079	{
2016		- phys_addr_t pfn;
2017		- int err;
2018		-
2019		- if (vma->vm_end - vma->vm_start != PAGE_SIZE)
	2080	+ if ((vma->vm_end - vma->vm_start != PAGE_SIZE) \|\|
	2081	+ !(vma->vm_flags & VM_SHARED))
2020	2082	return -EINVAL;
2021	2083
2022	2084	if (get_index(vma->vm_pgoff) != MLX5_IB_CLOCK_INFO_V1)
2023	2085	return -EOPNOTSUPP;
2024	2086
2025		- if (vma->vm_flags & VM_WRITE)
	2087	+ if (vma->vm_flags & (VM_WRITE \| VM_EXEC))
2026	2088	return -EPERM;
2027	2089	vma->vm_flags &= ~VM_MAYWRITE;
2028	2090
2029		- if (!dev->mdev->clock_info_page)
	2091	+ if (!dev->mdev->clock_info)
2030	2092	return -EOPNOTSUPP;
2031	2093
2032		- pfn = page_to_pfn(dev->mdev->clock_info_page);
2033		- err = remap_pfn_range(vma, vma->vm_start, pfn, PAGE_SIZE,
2034		- vma->vm_page_prot);
2035		- if (err)
2036		- return err;
	2094	+ return vm_insert_page(vma, vma->vm_start,
	2095	+ virt_to_page(dev->mdev->clock_info));
	2096	+}
2037	2097
2038		- return mlx5_ib_set_vma_data(vma, context);
	2098	+static void mlx5_ib_mmap_free(struct rdma_user_mmap_entry *entry)
	2099	+{
	2100	+ struct mlx5_user_mmap_entry *mentry = to_mmmap(entry);
	2101	+ struct mlx5_ib_dev *dev = to_mdev(entry->ucontext->device);
	2102	+ struct mlx5_var_table *var_table = &dev->var_table;
	2103	+ struct mlx5_ib_dm *mdm;
	2104	+
	2105	+ switch (mentry->mmap_flag) {
	2106	+ case MLX5_IB_MMAP_TYPE_MEMIC:
	2107	+ mdm = container_of(mentry, struct mlx5_ib_dm, mentry);
	2108	+ mlx5_cmd_dealloc_memic(&dev->dm, mdm->dev_addr,
	2109	+ mdm->size);
	2110	+ kfree(mdm);
	2111	+ break;
	2112	+ case MLX5_IB_MMAP_TYPE_VAR:
	2113	+ mutex_lock(&var_table->bitmap_lock);
	2114	+ clear_bit(mentry->page_idx, var_table->bitmap);
	2115	+ mutex_unlock(&var_table->bitmap_lock);
	2116	+ kfree(mentry);
	2117	+ break;
	2118	+ case MLX5_IB_MMAP_TYPE_UAR_WC:
	2119	+ case MLX5_IB_MMAP_TYPE_UAR_NC:
	2120	+ mlx5_cmd_free_uar(dev->mdev, mentry->page_idx);
	2121	+ kfree(mentry);
	2122	+ break;
	2123	+ default:
	2124	+ WARN_ON(true);
	2125	+ }
2039	2126	}
2040	2127
2041	2128	static int uar_mmap(struct mlx5_ib_dev *dev, enum mlx5_ib_mmap_cmd cmd,
..	..	@@ -2052,6 +2139,9 @@
2052	2139	int dyn_uar = (cmd == MLX5_IB_MMAP_ALLOC_WC);
2053	2140	int max_valid_idx = dyn_uar ? bfregi->num_sys_pages :
2054	2141	bfregi->num_static_sys_pages;
	2142	+
	2143	+ if (bfregi->lib_uar_dyn)
	2144	+ return -EINVAL;
2055	2145
2056	2146	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
2057	2147	return -EINVAL;
..	..	@@ -2070,14 +2160,6 @@
2070	2160	switch (cmd) {
2071	2161	case MLX5_IB_MMAP_WC_PAGE:
2072	2162	case MLX5_IB_MMAP_ALLOC_WC:
2073		-/* Some architectures don't support WC memory */
2074		-#if defined(CONFIG_X86)
2075		- if (!pat_enabled())
2076		- return -EPERM;
2077		-#elif !(defined(CONFIG_PPC) \|\| (defined(CONFIG_ARM) && defined(CONFIG_MMU)))
2078		- return -EPERM;
2079		-#endif
2080		- /* fall through */
2081	2163	case MLX5_IB_MMAP_REGULAR_PAGE:
2082	2164	/* For MLX5_IB_MMAP_REGULAR_PAGE do the best effort to get WC */
2083	2165	prot = pgprot_writecombine(vma->vm_page_prot);
..	..	@@ -2125,20 +2207,14 @@
2125	2207	pfn = uar_index2pfn(dev, uar_index);
2126	2208	mlx5_ib_dbg(dev, "uar idx 0x%lx, pfn %pa\n", idx, &pfn);
2127	2209
2128		- vma->vm_page_prot = prot;
2129		- err = io_remap_pfn_range(vma, vma->vm_start, pfn,
2130		- PAGE_SIZE, vma->vm_page_prot);
	2210	+ err = rdma_user_mmap_io(&context->ibucontext, vma, pfn, PAGE_SIZE,
	2211	+ prot, NULL);
2131	2212	if (err) {
2132	2213	mlx5_ib_err(dev,
2133		- "io_remap_pfn_range failed with error=%d, mmap_cmd=%s\n",
	2214	+ "rdma_user_mmap_io failed with error=%d, mmap_cmd=%s\n",
2134	2215	err, mmap_cmd2str(cmd));
2135		- err = -EAGAIN;
2136	2216	goto err;
2137	2217	}
2138		-
2139		- err = mlx5_ib_set_vma_data(vma, context);
2140		- if (err)
2141		- goto err;
2142	2218
2143	2219	if (dyn_uar)
2144	2220	bfregi->sys_pages[idx] = uar_index;
..	..	@@ -2156,32 +2232,68 @@
2156	2232	return err;
2157	2233	}
2158	2234
2159		-static int dm_mmap(struct ib_ucontext context, struct vm_area_struct vma)
	2235	+static int add_dm_mmap_entry(struct ib_ucontext *context,
	2236	+ struct mlx5_ib_dm *mdm,
	2237	+ u64 address)
2160	2238	{
2161		- struct mlx5_ib_ucontext *mctx = to_mucontext(context);
2162		- struct mlx5_ib_dev *dev = to_mdev(context->device);
2163		- u16 page_idx = get_extended_index(vma->vm_pgoff);
2164		- size_t map_size = vma->vm_end - vma->vm_start;
2165		- u32 npages = map_size >> PAGE_SHIFT;
2166		- phys_addr_t pfn;
2167		- pgprot_t prot;
	2239	+ mdm->mentry.mmap_flag = MLX5_IB_MMAP_TYPE_MEMIC;
	2240	+ mdm->mentry.address = address;
	2241	+ return rdma_user_mmap_entry_insert_range(
	2242	+ context, &mdm->mentry.rdma_entry,
	2243	+ mdm->size,
	2244	+ MLX5_IB_MMAP_DEVICE_MEM << 16,
	2245	+ (MLX5_IB_MMAP_DEVICE_MEM << 16) + (1UL << 16) - 1);
	2246	+}
2168	2247
2169		- if (find_next_zero_bit(mctx->dm_pages, page_idx + npages, page_idx) !=
2170		- page_idx + npages)
	2248	+static unsigned long mlx5_vma_to_pgoff(struct vm_area_struct *vma)
	2249	+{
	2250	+ unsigned long idx;
	2251	+ u8 command;
	2252	+
	2253	+ command = get_command(vma->vm_pgoff);
	2254	+ idx = get_extended_index(vma->vm_pgoff);
	2255	+
	2256	+ return (command << 16 \| idx);
	2257	+}
	2258	+
	2259	+static int mlx5_ib_mmap_offset(struct mlx5_ib_dev *dev,
	2260	+ struct vm_area_struct *vma,
	2261	+ struct ib_ucontext *ucontext)
	2262	+{
	2263	+ struct mlx5_user_mmap_entry *mentry;
	2264	+ struct rdma_user_mmap_entry *entry;
	2265	+ unsigned long pgoff;
	2266	+ pgprot_t prot;
	2267	+ phys_addr_t pfn;
	2268	+ int ret;
	2269	+
	2270	+ pgoff = mlx5_vma_to_pgoff(vma);
	2271	+ entry = rdma_user_mmap_entry_get_pgoff(ucontext, pgoff);
	2272	+ if (!entry)
2171	2273	return -EINVAL;
2172	2274
2173		- pfn = ((pci_resource_start(dev->mdev->pdev, 0) +
2174		- MLX5_CAP64_DEV_MEM(dev->mdev, memic_bar_start_addr)) >>
2175		- PAGE_SHIFT) +
2176		- page_idx;
2177		- prot = pgprot_writecombine(vma->vm_page_prot);
2178		- vma->vm_page_prot = prot;
	2275	+ mentry = to_mmmap(entry);
	2276	+ pfn = (mentry->address >> PAGE_SHIFT);
	2277	+ if (mentry->mmap_flag == MLX5_IB_MMAP_TYPE_VAR \|\|
	2278	+ mentry->mmap_flag == MLX5_IB_MMAP_TYPE_UAR_NC)
	2279	+ prot = pgprot_noncached(vma->vm_page_prot);
	2280	+ else
	2281	+ prot = pgprot_writecombine(vma->vm_page_prot);
	2282	+ ret = rdma_user_mmap_io(ucontext, vma, pfn,
	2283	+ entry->npages * PAGE_SIZE,
	2284	+ prot,
	2285	+ entry);
	2286	+ rdma_user_mmap_entry_put(&mentry->rdma_entry);
	2287	+ return ret;
	2288	+}
2179	2289
2180		- if (io_remap_pfn_range(vma, vma->vm_start, pfn, map_size,
2181		- vma->vm_page_prot))
2182		- return -EAGAIN;
	2290	+static u64 mlx5_entry_to_mmap_offset(struct mlx5_user_mmap_entry *entry)
	2291	+{
	2292	+ u64 cmd = (entry->rdma_entry.start_pgoff >> 16) & 0xFFFF;
	2293	+ u64 index = entry->rdma_entry.start_pgoff & 0xFFFF;
2183	2294
2184		- return mlx5_ib_set_vma_data(vma, mctx);
	2295	+ return (((index >> 8) << 16) \| (cmd << MLX5_IB_MMAP_CMD_SHIFT) \|
	2296	+ (index & 0xFF)) << PAGE_SHIFT;
2185	2297	}
2186	2298
2187	2299	static int mlx5_ib_mmap(struct ib_ucontext ibcontext, struct vm_area_struct vma)
..	..	@@ -2194,9 +2306,12 @@
2194	2306	command = get_command(vma->vm_pgoff);
2195	2307	switch (command) {
2196	2308	case MLX5_IB_MMAP_WC_PAGE:
	2309	+ case MLX5_IB_MMAP_ALLOC_WC:
	2310	+ if (!dev->wc_support)
	2311	+ return -EPERM;
	2312	+ fallthrough;
2197	2313	case MLX5_IB_MMAP_NC_PAGE:
2198	2314	case MLX5_IB_MMAP_REGULAR_PAGE:
2199		- case MLX5_IB_MMAP_ALLOC_WC:
2200	2315	return uar_mmap(dev, command, vma, context);
2201	2316
2202	2317	case MLX5_IB_MMAP_GET_CONTIGUOUS_PAGES:
..	..	@@ -2214,25 +2329,128 @@
2214	2329	if (PAGE_SIZE > 4096)
2215	2330	return -EOPNOTSUPP;
2216	2331
2217		- vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
2218	2332	pfn = (dev->mdev->iseg_base +
2219	2333	offsetof(struct mlx5_init_seg, internal_timer_h)) >>
2220	2334	PAGE_SHIFT;
2221		- if (io_remap_pfn_range(vma, vma->vm_start, pfn,
2222		- PAGE_SIZE, vma->vm_page_prot))
2223		- return -EAGAIN;
2224		- break;
	2335	+ return rdma_user_mmap_io(&context->ibucontext, vma, pfn,
	2336	+ PAGE_SIZE,
	2337	+ pgprot_noncached(vma->vm_page_prot),
	2338	+ NULL);
2225	2339	case MLX5_IB_MMAP_CLOCK_INFO:
2226	2340	return mlx5_ib_mmap_clock_info_page(dev, vma, context);
2227	2341
2228		- case MLX5_IB_MMAP_DEVICE_MEM:
2229		- return dm_mmap(ibcontext, vma);
2230		-
2231	2342	default:
2232		- return -EINVAL;
	2343	+ return mlx5_ib_mmap_offset(dev, vma, ibcontext);
2233	2344	}
2234	2345
2235	2346	return 0;
	2347	+}
	2348	+
	2349	+static inline int check_dm_type_support(struct mlx5_ib_dev *dev,
	2350	+ u32 type)
	2351	+{
	2352	+ switch (type) {
	2353	+ case MLX5_IB_UAPI_DM_TYPE_MEMIC:
	2354	+ if (!MLX5_CAP_DEV_MEM(dev->mdev, memic))
	2355	+ return -EOPNOTSUPP;
	2356	+ break;
	2357	+ case MLX5_IB_UAPI_DM_TYPE_STEERING_SW_ICM:
	2358	+ case MLX5_IB_UAPI_DM_TYPE_HEADER_MODIFY_SW_ICM:
	2359	+ if (!capable(CAP_SYS_RAWIO) \|\|
	2360	+ !capable(CAP_NET_RAW))
	2361	+ return -EPERM;
	2362	+
	2363	+ if (!(MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev, sw_owner) \|\|
	2364	+ MLX5_CAP_FLOWTABLE_NIC_TX(dev->mdev, sw_owner) \|\|
	2365	+ MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev, sw_owner_v2) \|\|
	2366	+ MLX5_CAP_FLOWTABLE_NIC_TX(dev->mdev, sw_owner_v2)))
	2367	+ return -EOPNOTSUPP;
	2368	+ break;
	2369	+ }
	2370	+
	2371	+ return 0;
	2372	+}
	2373	+
	2374	+static int handle_alloc_dm_memic(struct ib_ucontext *ctx,
	2375	+ struct mlx5_ib_dm *dm,
	2376	+ struct ib_dm_alloc_attr *attr,
	2377	+ struct uverbs_attr_bundle *attrs)
	2378	+{
	2379	+ struct mlx5_dm *dm_db = &to_mdev(ctx->device)->dm;
	2380	+ u64 start_offset;
	2381	+ u16 page_idx;
	2382	+ int err;
	2383	+ u64 address;
	2384	+
	2385	+ dm->size = roundup(attr->length, MLX5_MEMIC_BASE_SIZE);
	2386	+
	2387	+ err = mlx5_cmd_alloc_memic(dm_db, &dm->dev_addr,
	2388	+ dm->size, attr->alignment);
	2389	+ if (err)
	2390	+ return err;
	2391	+
	2392	+ address = dm->dev_addr & PAGE_MASK;
	2393	+ err = add_dm_mmap_entry(ctx, dm, address);
	2394	+ if (err)
	2395	+ goto err_dealloc;
	2396	+
	2397	+ page_idx = dm->mentry.rdma_entry.start_pgoff & 0xFFFF;
	2398	+ err = uverbs_copy_to(attrs,
	2399	+ MLX5_IB_ATTR_ALLOC_DM_RESP_PAGE_INDEX,
	2400	+ &page_idx,
	2401	+ sizeof(page_idx));
	2402	+ if (err)
	2403	+ goto err_copy;
	2404	+
	2405	+ start_offset = dm->dev_addr & ~PAGE_MASK;
	2406	+ err = uverbs_copy_to(attrs,
	2407	+ MLX5_IB_ATTR_ALLOC_DM_RESP_START_OFFSET,
	2408	+ &start_offset, sizeof(start_offset));
	2409	+ if (err)
	2410	+ goto err_copy;
	2411	+
	2412	+ return 0;
	2413	+
	2414	+err_copy:
	2415	+ rdma_user_mmap_entry_remove(&dm->mentry.rdma_entry);
	2416	+err_dealloc:
	2417	+ mlx5_cmd_dealloc_memic(dm_db, dm->dev_addr, dm->size);
	2418	+
	2419	+ return err;
	2420	+}
	2421	+
	2422	+static int handle_alloc_dm_sw_icm(struct ib_ucontext *ctx,
	2423	+ struct mlx5_ib_dm *dm,
	2424	+ struct ib_dm_alloc_attr *attr,
	2425	+ struct uverbs_attr_bundle *attrs,
	2426	+ int type)
	2427	+{
	2428	+ struct mlx5_core_dev *dev = to_mdev(ctx->device)->mdev;
	2429	+ u64 act_size;
	2430	+ int err;
	2431	+
	2432	+ /* Allocation size must a multiple of the basic block size
	2433	+ * and a power of 2.
	2434	+ */
	2435	+ act_size = round_up(attr->length, MLX5_SW_ICM_BLOCK_SIZE(dev));
	2436	+ act_size = roundup_pow_of_two(act_size);
	2437	+
	2438	+ dm->size = act_size;
	2439	+ err = mlx5_dm_sw_icm_alloc(dev, type, act_size, attr->alignment,
	2440	+ to_mucontext(ctx)->devx_uid, &dm->dev_addr,
	2441	+ &dm->icm_dm.obj_id);
	2442	+ if (err)
	2443	+ return err;
	2444	+
	2445	+ err = uverbs_copy_to(attrs,
	2446	+ MLX5_IB_ATTR_ALLOC_DM_RESP_START_OFFSET,
	2447	+ &dm->dev_addr, sizeof(dm->dev_addr));
	2448	+ if (err)
	2449	+ mlx5_dm_sw_icm_dealloc(dev, type, dm->size,
	2450	+ to_mucontext(ctx)->devx_uid, dm->dev_addr,
	2451	+ dm->icm_dm.obj_id);
	2452	+
	2453	+ return err;
2236	2454	}
2237	2455
2238	2456	struct ib_dm mlx5_ib_alloc_dm(struct ib_device ibdev,
..	..	@@ -2240,1793 +2458,132 @@
2240	2458	struct ib_dm_alloc_attr *attr,
2241	2459	struct uverbs_attr_bundle *attrs)
2242	2460	{
2243		- u64 act_size = roundup(attr->length, MLX5_MEMIC_BASE_SIZE);
2244		- struct mlx5_memic *memic = &to_mdev(ibdev)->memic;
2245		- phys_addr_t memic_addr;
2246	2461	struct mlx5_ib_dm *dm;
2247		- u64 start_offset;
2248		- u32 page_idx;
	2462	+ enum mlx5_ib_uapi_dm_type type;
2249	2463	int err;
	2464	+
	2465	+ err = uverbs_get_const_default(&type, attrs,
	2466	+ MLX5_IB_ATTR_ALLOC_DM_REQ_TYPE,
	2467	+ MLX5_IB_UAPI_DM_TYPE_MEMIC);
	2468	+ if (err)
	2469	+ return ERR_PTR(err);
	2470	+
	2471	+ mlx5_ib_dbg(to_mdev(ibdev), "alloc_dm req: dm_type=%d user_length=0x%llx log_alignment=%d\n",
	2472	+ type, attr->length, attr->alignment);
	2473	+
	2474	+ err = check_dm_type_support(to_mdev(ibdev), type);
	2475	+ if (err)
	2476	+ return ERR_PTR(err);
2250	2477
2251	2478	dm = kzalloc(sizeof(*dm), GFP_KERNEL);
2252	2479	if (!dm)
2253	2480	return ERR_PTR(-ENOMEM);
2254	2481
2255		- mlx5_ib_dbg(to_mdev(ibdev), "alloc_memic req: user_length=0x%llx act_length=0x%llx log_alignment=%d\n",
2256		- attr->length, act_size, attr->alignment);
	2482	+ dm->type = type;
2257	2483
2258		- err = mlx5_cmd_alloc_memic(memic, &memic_addr,
2259		- act_size, attr->alignment);
	2484	+ switch (type) {
	2485	+ case MLX5_IB_UAPI_DM_TYPE_MEMIC:
	2486	+ err = handle_alloc_dm_memic(context, dm,
	2487	+ attr,
	2488	+ attrs);
	2489	+ break;
	2490	+ case MLX5_IB_UAPI_DM_TYPE_STEERING_SW_ICM:
	2491	+ err = handle_alloc_dm_sw_icm(context, dm,
	2492	+ attr, attrs,
	2493	+ MLX5_SW_ICM_TYPE_STEERING);
	2494	+ break;
	2495	+ case MLX5_IB_UAPI_DM_TYPE_HEADER_MODIFY_SW_ICM:
	2496	+ err = handle_alloc_dm_sw_icm(context, dm,
	2497	+ attr, attrs,
	2498	+ MLX5_SW_ICM_TYPE_HEADER_MODIFY);
	2499	+ break;
	2500	+ default:
	2501	+ err = -EOPNOTSUPP;
	2502	+ }
	2503	+
2260	2504	if (err)
2261	2505	goto err_free;
2262	2506
2263		- start_offset = memic_addr & ~PAGE_MASK;
2264		- page_idx = (memic_addr - pci_resource_start(memic->dev->pdev, 0) -
2265		- MLX5_CAP64_DEV_MEM(memic->dev, memic_bar_start_addr)) >>
2266		- PAGE_SHIFT;
2267		-
2268		- err = uverbs_copy_to(attrs,
2269		- MLX5_IB_ATTR_ALLOC_DM_RESP_START_OFFSET,
2270		- &start_offset, sizeof(start_offset));
2271		- if (err)
2272		- goto err_dealloc;
2273		-
2274		- err = uverbs_copy_to(attrs,
2275		- MLX5_IB_ATTR_ALLOC_DM_RESP_PAGE_INDEX,
2276		- &page_idx, sizeof(page_idx));
2277		- if (err)
2278		- goto err_dealloc;
2279		-
2280		- bitmap_set(to_mucontext(context)->dm_pages, page_idx,
2281		- DIV_ROUND_UP(act_size, PAGE_SIZE));
2282		-
2283		- dm->dev_addr = memic_addr;
2284		-
2285	2507	return &dm->ibdm;
2286	2508
2287		-err_dealloc:
2288		- mlx5_cmd_dealloc_memic(memic, memic_addr,
2289		- act_size);
2290	2509	err_free:
2291	2510	kfree(dm);
2292	2511	return ERR_PTR(err);
2293	2512	}
2294	2513
2295		-int mlx5_ib_dealloc_dm(struct ib_dm *ibdm)
	2514	+int mlx5_ib_dealloc_dm(struct ib_dm ibdm, struct uverbs_attr_bundle attrs)
2296	2515	{
2297		- struct mlx5_memic *memic = &to_mdev(ibdm->device)->memic;
	2516	+ struct mlx5_ib_ucontext *ctx = rdma_udata_to_drv_context(
	2517	+ &attrs->driver_udata, struct mlx5_ib_ucontext, ibucontext);
	2518	+ struct mlx5_core_dev *dev = to_mdev(ibdm->device)->mdev;
2298	2519	struct mlx5_ib_dm *dm = to_mdm(ibdm);
2299		- u64 act_size = roundup(dm->ibdm.length, MLX5_MEMIC_BASE_SIZE);
2300		- u32 page_idx;
2301	2520	int ret;
2302	2521
2303		- ret = mlx5_cmd_dealloc_memic(memic, dm->dev_addr, act_size);
2304		- if (ret)
2305		- return ret;
2306		-
2307		- page_idx = (dm->dev_addr - pci_resource_start(memic->dev->pdev, 0) -
2308		- MLX5_CAP64_DEV_MEM(memic->dev, memic_bar_start_addr)) >>
2309		- PAGE_SHIFT;
2310		- bitmap_clear(to_mucontext(ibdm->uobject->context)->dm_pages,
2311		- page_idx,
2312		- DIV_ROUND_UP(act_size, PAGE_SIZE));
	2522	+ switch (dm->type) {
	2523	+ case MLX5_IB_UAPI_DM_TYPE_MEMIC:
	2524	+ rdma_user_mmap_entry_remove(&dm->mentry.rdma_entry);
	2525	+ return 0;
	2526	+ case MLX5_IB_UAPI_DM_TYPE_STEERING_SW_ICM:
	2527	+ ret = mlx5_dm_sw_icm_dealloc(dev, MLX5_SW_ICM_TYPE_STEERING,
	2528	+ dm->size, ctx->devx_uid, dm->dev_addr,
	2529	+ dm->icm_dm.obj_id);
	2530	+ if (ret)
	2531	+ return ret;
	2532	+ break;
	2533	+ case MLX5_IB_UAPI_DM_TYPE_HEADER_MODIFY_SW_ICM:
	2534	+ ret = mlx5_dm_sw_icm_dealloc(dev, MLX5_SW_ICM_TYPE_HEADER_MODIFY,
	2535	+ dm->size, ctx->devx_uid, dm->dev_addr,
	2536	+ dm->icm_dm.obj_id);
	2537	+ if (ret)
	2538	+ return ret;
	2539	+ break;
	2540	+ default:
	2541	+ return -EOPNOTSUPP;
	2542	+ }
2313	2543
2314	2544	kfree(dm);
2315	2545
2316	2546	return 0;
2317	2547	}
2318	2548
2319		-static struct ib_pd mlx5_ib_alloc_pd(struct ib_device ibdev,
2320		- struct ib_ucontext *context,
2321		- struct ib_udata *udata)
	2549	+static int mlx5_ib_alloc_pd(struct ib_pd ibpd, struct ib_udata udata)
2322	2550	{
	2551	+ struct mlx5_ib_pd *pd = to_mpd(ibpd);
	2552	+ struct ib_device *ibdev = ibpd->device;
2323	2553	struct mlx5_ib_alloc_pd_resp resp;
2324		- struct mlx5_ib_pd *pd;
2325	2554	int err;
	2555	+ u32 out[MLX5_ST_SZ_DW(alloc_pd_out)] = {};
	2556	+ u32 in[MLX5_ST_SZ_DW(alloc_pd_in)] = {};
	2557	+ u16 uid = 0;
	2558	+ struct mlx5_ib_ucontext *context = rdma_udata_to_drv_context(
	2559	+ udata, struct mlx5_ib_ucontext, ibucontext);
2326	2560
2327		- pd = kmalloc(sizeof(*pd), GFP_KERNEL);
2328		- if (!pd)
2329		- return ERR_PTR(-ENOMEM);
	2561	+ uid = context ? context->devx_uid : 0;
	2562	+ MLX5_SET(alloc_pd_in, in, opcode, MLX5_CMD_OP_ALLOC_PD);
	2563	+ MLX5_SET(alloc_pd_in, in, uid, uid);
	2564	+ err = mlx5_cmd_exec_inout(to_mdev(ibdev)->mdev, alloc_pd, in, out);
	2565	+ if (err)
	2566	+ return err;
2330	2567
2331		- err = mlx5_core_alloc_pd(to_mdev(ibdev)->mdev, &pd->pdn);
2332		- if (err) {
2333		- kfree(pd);
2334		- return ERR_PTR(err);
2335		- }
2336		-
2337		- if (context) {
	2568	+ pd->pdn = MLX5_GET(alloc_pd_out, out, pd);
	2569	+ pd->uid = uid;
	2570	+ if (udata) {
2338	2571	resp.pdn = pd->pdn;
2339	2572	if (ib_copy_to_udata(udata, &resp, sizeof(resp))) {
2340		- mlx5_core_dealloc_pd(to_mdev(ibdev)->mdev, pd->pdn);
2341		- kfree(pd);
2342		- return ERR_PTR(-EFAULT);
	2573	+ mlx5_cmd_dealloc_pd(to_mdev(ibdev)->mdev, pd->pdn, uid);
	2574	+ return -EFAULT;
2343	2575	}
2344	2576	}
2345	2577
2346		- return &pd->ibpd;
	2578	+ return 0;
2347	2579	}
2348	2580
2349		-static int mlx5_ib_dealloc_pd(struct ib_pd *pd)
	2581	+static int mlx5_ib_dealloc_pd(struct ib_pd pd, struct ib_udata udata)
2350	2582	{
2351	2583	struct mlx5_ib_dev *mdev = to_mdev(pd->device);
2352	2584	struct mlx5_ib_pd *mpd = to_mpd(pd);
2353	2585
2354		- mlx5_core_dealloc_pd(mdev->mdev, mpd->pdn);
2355		- kfree(mpd);
2356		-
2357		- return 0;
2358		-}
2359		-
2360		-enum {
2361		- MATCH_CRITERIA_ENABLE_OUTER_BIT,
2362		- MATCH_CRITERIA_ENABLE_MISC_BIT,
2363		- MATCH_CRITERIA_ENABLE_INNER_BIT,
2364		- MATCH_CRITERIA_ENABLE_MISC2_BIT
2365		-};
2366		-
2367		-#define HEADER_IS_ZERO(match_criteria, headers) \
2368		- !(memchr_inv(MLX5_ADDR_OF(fte_match_param, match_criteria, headers), \
2369		- 0, MLX5_FLD_SZ_BYTES(fte_match_param, headers))) \
2370		-
2371		-static u8 get_match_criteria_enable(u32 *match_criteria)
2372		-{
2373		- u8 match_criteria_enable;
2374		-
2375		- match_criteria_enable =
2376		- (!HEADER_IS_ZERO(match_criteria, outer_headers)) <<
2377		- MATCH_CRITERIA_ENABLE_OUTER_BIT;
2378		- match_criteria_enable \|=
2379		- (!HEADER_IS_ZERO(match_criteria, misc_parameters)) <<
2380		- MATCH_CRITERIA_ENABLE_MISC_BIT;
2381		- match_criteria_enable \|=
2382		- (!HEADER_IS_ZERO(match_criteria, inner_headers)) <<
2383		- MATCH_CRITERIA_ENABLE_INNER_BIT;
2384		- match_criteria_enable \|=
2385		- (!HEADER_IS_ZERO(match_criteria, misc_parameters_2)) <<
2386		- MATCH_CRITERIA_ENABLE_MISC2_BIT;
2387		-
2388		- return match_criteria_enable;
2389		-}
2390		-
2391		-static int set_proto(void outer_c, void outer_v, u8 mask, u8 val)
2392		-{
2393		- u8 entry_mask;
2394		- u8 entry_val;
2395		- int err = 0;
2396		-
2397		- if (!mask)
2398		- goto out;
2399		-
2400		- entry_mask = MLX5_GET(fte_match_set_lyr_2_4, outer_c,
2401		- ip_protocol);
2402		- entry_val = MLX5_GET(fte_match_set_lyr_2_4, outer_v,
2403		- ip_protocol);
2404		- if (!entry_mask) {
2405		- MLX5_SET(fte_match_set_lyr_2_4, outer_c, ip_protocol, mask);
2406		- MLX5_SET(fte_match_set_lyr_2_4, outer_v, ip_protocol, val);
2407		- goto out;
2408		- }
2409		- /* Don't override existing ip protocol */
2410		- if (mask != entry_mask \|\| val != entry_val)
2411		- err = -EINVAL;
2412		-out:
2413		- return err;
2414		-}
2415		-
2416		-static void set_flow_label(void misc_c, void misc_v, u32 mask, u32 val,
2417		- bool inner)
2418		-{
2419		- if (inner) {
2420		- MLX5_SET(fte_match_set_misc,
2421		- misc_c, inner_ipv6_flow_label, mask);
2422		- MLX5_SET(fte_match_set_misc,
2423		- misc_v, inner_ipv6_flow_label, val);
2424		- } else {
2425		- MLX5_SET(fte_match_set_misc,
2426		- misc_c, outer_ipv6_flow_label, mask);
2427		- MLX5_SET(fte_match_set_misc,
2428		- misc_v, outer_ipv6_flow_label, val);
2429		- }
2430		-}
2431		-
2432		-static void set_tos(void outer_c, void outer_v, u8 mask, u8 val)
2433		-{
2434		- MLX5_SET(fte_match_set_lyr_2_4, outer_c, ip_ecn, mask);
2435		- MLX5_SET(fte_match_set_lyr_2_4, outer_v, ip_ecn, val);
2436		- MLX5_SET(fte_match_set_lyr_2_4, outer_c, ip_dscp, mask >> 2);
2437		- MLX5_SET(fte_match_set_lyr_2_4, outer_v, ip_dscp, val >> 2);
2438		-}
2439		-
2440		-static int check_mpls_supp_fields(u32 field_support, const __be32 *set_mask)
2441		-{
2442		- if (MLX5_GET(fte_match_mpls, set_mask, mpls_label) &&
2443		- !(field_support & MLX5_FIELD_SUPPORT_MPLS_LABEL))
2444		- return -EOPNOTSUPP;
2445		-
2446		- if (MLX5_GET(fte_match_mpls, set_mask, mpls_exp) &&
2447		- !(field_support & MLX5_FIELD_SUPPORT_MPLS_EXP))
2448		- return -EOPNOTSUPP;
2449		-
2450		- if (MLX5_GET(fte_match_mpls, set_mask, mpls_s_bos) &&
2451		- !(field_support & MLX5_FIELD_SUPPORT_MPLS_S_BOS))
2452		- return -EOPNOTSUPP;
2453		-
2454		- if (MLX5_GET(fte_match_mpls, set_mask, mpls_ttl) &&
2455		- !(field_support & MLX5_FIELD_SUPPORT_MPLS_TTL))
2456		- return -EOPNOTSUPP;
2457		-
2458		- return 0;
2459		-}
2460		-
2461		-#define LAST_ETH_FIELD vlan_tag
2462		-#define LAST_IB_FIELD sl
2463		-#define LAST_IPV4_FIELD tos
2464		-#define LAST_IPV6_FIELD traffic_class
2465		-#define LAST_TCP_UDP_FIELD src_port
2466		-#define LAST_TUNNEL_FIELD tunnel_id
2467		-#define LAST_FLOW_TAG_FIELD tag_id
2468		-#define LAST_DROP_FIELD size
2469		-#define LAST_COUNTERS_FIELD counters
2470		-
2471		-/* Field is the last supported field */
2472		-#define FIELDS_NOT_SUPPORTED(filter, field)\
2473		- memchr_inv((void *)&filter.field +\
2474		- sizeof(filter.field), 0,\
2475		- sizeof(filter) -\
2476		- offsetof(typeof(filter), field) -\
2477		- sizeof(filter.field))
2478		-
2479		-static int parse_flow_flow_action(const union ib_flow_spec *ib_spec,
2480		- const struct ib_flow_attr *flow_attr,
2481		- struct mlx5_flow_act *action)
2482		-{
2483		- struct mlx5_ib_flow_action *maction = to_mflow_act(ib_spec->action.act);
2484		-
2485		- switch (maction->ib_action.type) {
2486		- case IB_FLOW_ACTION_ESP:
2487		- /* Currently only AES_GCM keymat is supported by the driver */
2488		- action->esp_id = (uintptr_t)maction->esp_aes_gcm.ctx;
2489		- action->action \|= flow_attr->flags & IB_FLOW_ATTR_FLAGS_EGRESS ?
2490		- MLX5_FLOW_CONTEXT_ACTION_ENCRYPT :
2491		- MLX5_FLOW_CONTEXT_ACTION_DECRYPT;
2492		- return 0;
2493		- default:
2494		- return -EOPNOTSUPP;
2495		- }
2496		-}
2497		-
2498		-static int parse_flow_attr(struct mlx5_core_dev mdev, u32 match_c,
2499		- u32 match_v, const union ib_flow_spec ib_spec,
2500		- const struct ib_flow_attr *flow_attr,
2501		- struct mlx5_flow_act *action, u32 prev_type)
2502		-{
2503		- void *misc_params_c = MLX5_ADDR_OF(fte_match_param, match_c,
2504		- misc_parameters);
2505		- void *misc_params_v = MLX5_ADDR_OF(fte_match_param, match_v,
2506		- misc_parameters);
2507		- void *misc_params2_c = MLX5_ADDR_OF(fte_match_param, match_c,
2508		- misc_parameters_2);
2509		- void *misc_params2_v = MLX5_ADDR_OF(fte_match_param, match_v,
2510		- misc_parameters_2);
2511		- void *headers_c;
2512		- void *headers_v;
2513		- int match_ipv;
2514		- int ret;
2515		-
2516		- if (ib_spec->type & IB_FLOW_SPEC_INNER) {
2517		- headers_c = MLX5_ADDR_OF(fte_match_param, match_c,
2518		- inner_headers);
2519		- headers_v = MLX5_ADDR_OF(fte_match_param, match_v,
2520		- inner_headers);
2521		- match_ipv = MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
2522		- ft_field_support.inner_ip_version);
2523		- } else {
2524		- headers_c = MLX5_ADDR_OF(fte_match_param, match_c,
2525		- outer_headers);
2526		- headers_v = MLX5_ADDR_OF(fte_match_param, match_v,
2527		- outer_headers);
2528		- match_ipv = MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
2529		- ft_field_support.outer_ip_version);
2530		- }
2531		-
2532		- switch (ib_spec->type & ~IB_FLOW_SPEC_INNER) {
2533		- case IB_FLOW_SPEC_ETH:
2534		- if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
2535		- return -EOPNOTSUPP;
2536		-
2537		- ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2538		- dmac_47_16),
2539		- ib_spec->eth.mask.dst_mac);
2540		- ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2541		- dmac_47_16),
2542		- ib_spec->eth.val.dst_mac);
2543		-
2544		- ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2545		- smac_47_16),
2546		- ib_spec->eth.mask.src_mac);
2547		- ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2548		- smac_47_16),
2549		- ib_spec->eth.val.src_mac);
2550		-
2551		- if (ib_spec->eth.mask.vlan_tag) {
2552		- MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2553		- cvlan_tag, 1);
2554		- MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2555		- cvlan_tag, 1);
2556		-
2557		- MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2558		- first_vid, ntohs(ib_spec->eth.mask.vlan_tag));
2559		- MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2560		- first_vid, ntohs(ib_spec->eth.val.vlan_tag));
2561		-
2562		- MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2563		- first_cfi,
2564		- ntohs(ib_spec->eth.mask.vlan_tag) >> 12);
2565		- MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2566		- first_cfi,
2567		- ntohs(ib_spec->eth.val.vlan_tag) >> 12);
2568		-
2569		- MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2570		- first_prio,
2571		- ntohs(ib_spec->eth.mask.vlan_tag) >> 13);
2572		- MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2573		- first_prio,
2574		- ntohs(ib_spec->eth.val.vlan_tag) >> 13);
2575		- }
2576		- MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2577		- ethertype, ntohs(ib_spec->eth.mask.ether_type));
2578		- MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2579		- ethertype, ntohs(ib_spec->eth.val.ether_type));
2580		- break;
2581		- case IB_FLOW_SPEC_IPV4:
2582		- if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
2583		- return -EOPNOTSUPP;
2584		-
2585		- if (match_ipv) {
2586		- MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2587		- ip_version, 0xf);
2588		- MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2589		- ip_version, MLX5_FS_IPV4_VERSION);
2590		- } else {
2591		- MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2592		- ethertype, 0xffff);
2593		- MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2594		- ethertype, ETH_P_IP);
2595		- }
2596		-
2597		- memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2598		- src_ipv4_src_ipv6.ipv4_layout.ipv4),
2599		- &ib_spec->ipv4.mask.src_ip,
2600		- sizeof(ib_spec->ipv4.mask.src_ip));
2601		- memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2602		- src_ipv4_src_ipv6.ipv4_layout.ipv4),
2603		- &ib_spec->ipv4.val.src_ip,
2604		- sizeof(ib_spec->ipv4.val.src_ip));
2605		- memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2606		- dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
2607		- &ib_spec->ipv4.mask.dst_ip,
2608		- sizeof(ib_spec->ipv4.mask.dst_ip));
2609		- memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2610		- dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
2611		- &ib_spec->ipv4.val.dst_ip,
2612		- sizeof(ib_spec->ipv4.val.dst_ip));
2613		-
2614		- set_tos(headers_c, headers_v,
2615		- ib_spec->ipv4.mask.tos, ib_spec->ipv4.val.tos);
2616		-
2617		- if (set_proto(headers_c, headers_v,
2618		- ib_spec->ipv4.mask.proto,
2619		- ib_spec->ipv4.val.proto))
2620		- return -EINVAL;
2621		- break;
2622		- case IB_FLOW_SPEC_IPV6:
2623		- if (FIELDS_NOT_SUPPORTED(ib_spec->ipv6.mask, LAST_IPV6_FIELD))
2624		- return -EOPNOTSUPP;
2625		-
2626		- if (match_ipv) {
2627		- MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2628		- ip_version, 0xf);
2629		- MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2630		- ip_version, MLX5_FS_IPV6_VERSION);
2631		- } else {
2632		- MLX5_SET(fte_match_set_lyr_2_4, headers_c,
2633		- ethertype, 0xffff);
2634		- MLX5_SET(fte_match_set_lyr_2_4, headers_v,
2635		- ethertype, ETH_P_IPV6);
2636		- }
2637		-
2638		- memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2639		- src_ipv4_src_ipv6.ipv6_layout.ipv6),
2640		- &ib_spec->ipv6.mask.src_ip,
2641		- sizeof(ib_spec->ipv6.mask.src_ip));
2642		- memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2643		- src_ipv4_src_ipv6.ipv6_layout.ipv6),
2644		- &ib_spec->ipv6.val.src_ip,
2645		- sizeof(ib_spec->ipv6.val.src_ip));
2646		- memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
2647		- dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
2648		- &ib_spec->ipv6.mask.dst_ip,
2649		- sizeof(ib_spec->ipv6.mask.dst_ip));
2650		- memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
2651		- dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
2652		- &ib_spec->ipv6.val.dst_ip,
2653		- sizeof(ib_spec->ipv6.val.dst_ip));
2654		-
2655		- set_tos(headers_c, headers_v,
2656		- ib_spec->ipv6.mask.traffic_class,
2657		- ib_spec->ipv6.val.traffic_class);
2658		-
2659		- if (set_proto(headers_c, headers_v,
2660		- ib_spec->ipv6.mask.next_hdr,
2661		- ib_spec->ipv6.val.next_hdr))
2662		- return -EINVAL;
2663		-
2664		- set_flow_label(misc_params_c, misc_params_v,
2665		- ntohl(ib_spec->ipv6.mask.flow_label),
2666		- ntohl(ib_spec->ipv6.val.flow_label),
2667		- ib_spec->type & IB_FLOW_SPEC_INNER);
2668		- break;
2669		- case IB_FLOW_SPEC_ESP:
2670		- if (ib_spec->esp.mask.seq)
2671		- return -EOPNOTSUPP;
2672		-
2673		- MLX5_SET(fte_match_set_misc, misc_params_c, outer_esp_spi,
2674		- ntohl(ib_spec->esp.mask.spi));
2675		- MLX5_SET(fte_match_set_misc, misc_params_v, outer_esp_spi,
2676		- ntohl(ib_spec->esp.val.spi));
2677		- break;
2678		- case IB_FLOW_SPEC_TCP:
2679		- if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask,
2680		- LAST_TCP_UDP_FIELD))
2681		- return -EOPNOTSUPP;
2682		-
2683		- if (set_proto(headers_c, headers_v, 0xff, IPPROTO_TCP))
2684		- return -EINVAL;
2685		-
2686		- MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_sport,
2687		- ntohs(ib_spec->tcp_udp.mask.src_port));
2688		- MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_sport,
2689		- ntohs(ib_spec->tcp_udp.val.src_port));
2690		-
2691		- MLX5_SET(fte_match_set_lyr_2_4, headers_c, tcp_dport,
2692		- ntohs(ib_spec->tcp_udp.mask.dst_port));
2693		- MLX5_SET(fte_match_set_lyr_2_4, headers_v, tcp_dport,
2694		- ntohs(ib_spec->tcp_udp.val.dst_port));
2695		- break;
2696		- case IB_FLOW_SPEC_UDP:
2697		- if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask,
2698		- LAST_TCP_UDP_FIELD))
2699		- return -EOPNOTSUPP;
2700		-
2701		- if (set_proto(headers_c, headers_v, 0xff, IPPROTO_UDP))
2702		- return -EINVAL;
2703		-
2704		- MLX5_SET(fte_match_set_lyr_2_4, headers_c, udp_sport,
2705		- ntohs(ib_spec->tcp_udp.mask.src_port));
2706		- MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_sport,
2707		- ntohs(ib_spec->tcp_udp.val.src_port));
2708		-
2709		- MLX5_SET(fte_match_set_lyr_2_4, headers_c, udp_dport,
2710		- ntohs(ib_spec->tcp_udp.mask.dst_port));
2711		- MLX5_SET(fte_match_set_lyr_2_4, headers_v, udp_dport,
2712		- ntohs(ib_spec->tcp_udp.val.dst_port));
2713		- break;
2714		- case IB_FLOW_SPEC_GRE:
2715		- if (ib_spec->gre.mask.c_ks_res0_ver)
2716		- return -EOPNOTSUPP;
2717		-
2718		- if (set_proto(headers_c, headers_v, 0xff, IPPROTO_GRE))
2719		- return -EINVAL;
2720		-
2721		- MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol,
2722		- 0xff);
2723		- MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
2724		- IPPROTO_GRE);
2725		-
2726		- MLX5_SET(fte_match_set_misc, misc_params_c, gre_protocol,
2727		- ntohs(ib_spec->gre.mask.protocol));
2728		- MLX5_SET(fte_match_set_misc, misc_params_v, gre_protocol,
2729		- ntohs(ib_spec->gre.val.protocol));
2730		-
2731		- memcpy(MLX5_ADDR_OF(fte_match_set_misc, misc_params_c,
2732		- gre_key_h),
2733		- &ib_spec->gre.mask.key,
2734		- sizeof(ib_spec->gre.mask.key));
2735		- memcpy(MLX5_ADDR_OF(fte_match_set_misc, misc_params_v,
2736		- gre_key_h),
2737		- &ib_spec->gre.val.key,
2738		- sizeof(ib_spec->gre.val.key));
2739		- break;
2740		- case IB_FLOW_SPEC_MPLS:
2741		- switch (prev_type) {
2742		- case IB_FLOW_SPEC_UDP:
2743		- if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
2744		- ft_field_support.outer_first_mpls_over_udp),
2745		- &ib_spec->mpls.mask.tag))
2746		- return -EOPNOTSUPP;
2747		-
2748		- memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
2749		- outer_first_mpls_over_udp),
2750		- &ib_spec->mpls.val.tag,
2751		- sizeof(ib_spec->mpls.val.tag));
2752		- memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
2753		- outer_first_mpls_over_udp),
2754		- &ib_spec->mpls.mask.tag,
2755		- sizeof(ib_spec->mpls.mask.tag));
2756		- break;
2757		- case IB_FLOW_SPEC_GRE:
2758		- if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
2759		- ft_field_support.outer_first_mpls_over_gre),
2760		- &ib_spec->mpls.mask.tag))
2761		- return -EOPNOTSUPP;
2762		-
2763		- memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
2764		- outer_first_mpls_over_gre),
2765		- &ib_spec->mpls.val.tag,
2766		- sizeof(ib_spec->mpls.val.tag));
2767		- memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
2768		- outer_first_mpls_over_gre),
2769		- &ib_spec->mpls.mask.tag,
2770		- sizeof(ib_spec->mpls.mask.tag));
2771		- break;
2772		- default:
2773		- if (ib_spec->type & IB_FLOW_SPEC_INNER) {
2774		- if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
2775		- ft_field_support.inner_first_mpls),
2776		- &ib_spec->mpls.mask.tag))
2777		- return -EOPNOTSUPP;
2778		-
2779		- memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
2780		- inner_first_mpls),
2781		- &ib_spec->mpls.val.tag,
2782		- sizeof(ib_spec->mpls.val.tag));
2783		- memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
2784		- inner_first_mpls),
2785		- &ib_spec->mpls.mask.tag,
2786		- sizeof(ib_spec->mpls.mask.tag));
2787		- } else {
2788		- if (check_mpls_supp_fields(MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
2789		- ft_field_support.outer_first_mpls),
2790		- &ib_spec->mpls.mask.tag))
2791		- return -EOPNOTSUPP;
2792		-
2793		- memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_v,
2794		- outer_first_mpls),
2795		- &ib_spec->mpls.val.tag,
2796		- sizeof(ib_spec->mpls.val.tag));
2797		- memcpy(MLX5_ADDR_OF(fte_match_set_misc2, misc_params2_c,
2798		- outer_first_mpls),
2799		- &ib_spec->mpls.mask.tag,
2800		- sizeof(ib_spec->mpls.mask.tag));
2801		- }
2802		- }
2803		- break;
2804		- case IB_FLOW_SPEC_VXLAN_TUNNEL:
2805		- if (FIELDS_NOT_SUPPORTED(ib_spec->tunnel.mask,
2806		- LAST_TUNNEL_FIELD))
2807		- return -EOPNOTSUPP;
2808		-
2809		- MLX5_SET(fte_match_set_misc, misc_params_c, vxlan_vni,
2810		- ntohl(ib_spec->tunnel.mask.tunnel_id));
2811		- MLX5_SET(fte_match_set_misc, misc_params_v, vxlan_vni,
2812		- ntohl(ib_spec->tunnel.val.tunnel_id));
2813		- break;
2814		- case IB_FLOW_SPEC_ACTION_TAG:
2815		- if (FIELDS_NOT_SUPPORTED(ib_spec->flow_tag,
2816		- LAST_FLOW_TAG_FIELD))
2817		- return -EOPNOTSUPP;
2818		- if (ib_spec->flow_tag.tag_id >= BIT(24))
2819		- return -EINVAL;
2820		-
2821		- action->flow_tag = ib_spec->flow_tag.tag_id;
2822		- action->has_flow_tag = true;
2823		- break;
2824		- case IB_FLOW_SPEC_ACTION_DROP:
2825		- if (FIELDS_NOT_SUPPORTED(ib_spec->drop,
2826		- LAST_DROP_FIELD))
2827		- return -EOPNOTSUPP;
2828		- action->action \|= MLX5_FLOW_CONTEXT_ACTION_DROP;
2829		- break;
2830		- case IB_FLOW_SPEC_ACTION_HANDLE:
2831		- ret = parse_flow_flow_action(ib_spec, flow_attr, action);
2832		- if (ret)
2833		- return ret;
2834		- break;
2835		- case IB_FLOW_SPEC_ACTION_COUNT:
2836		- if (FIELDS_NOT_SUPPORTED(ib_spec->flow_count,
2837		- LAST_COUNTERS_FIELD))
2838		- return -EOPNOTSUPP;
2839		-
2840		- /* for now support only one counters spec per flow */
2841		- if (action->action & MLX5_FLOW_CONTEXT_ACTION_COUNT)
2842		- return -EINVAL;
2843		-
2844		- action->counters = ib_spec->flow_count.counters;
2845		- action->action \|= MLX5_FLOW_CONTEXT_ACTION_COUNT;
2846		- break;
2847		- default:
2848		- return -EINVAL;
2849		- }
2850		-
2851		- return 0;
2852		-}
2853		-
2854		-/* If a flow could catch both multicast and unicast packets,
2855		- * it won't fall into the multicast flow steering table and this rule
2856		- * could steal other multicast packets.
2857		- */
2858		-static bool flow_is_multicast_only(const struct ib_flow_attr *ib_attr)
2859		-{
2860		- union ib_flow_spec *flow_spec;
2861		-
2862		- if (ib_attr->type != IB_FLOW_ATTR_NORMAL \|\|
2863		- ib_attr->num_of_specs < 1)
2864		- return false;
2865		-
2866		- flow_spec = (union ib_flow_spec *)(ib_attr + 1);
2867		- if (flow_spec->type == IB_FLOW_SPEC_IPV4) {
2868		- struct ib_flow_spec_ipv4 *ipv4_spec;
2869		-
2870		- ipv4_spec = (struct ib_flow_spec_ipv4 *)flow_spec;
2871		- if (ipv4_is_multicast(ipv4_spec->val.dst_ip))
2872		- return true;
2873		-
2874		- return false;
2875		- }
2876		-
2877		- if (flow_spec->type == IB_FLOW_SPEC_ETH) {
2878		- struct ib_flow_spec_eth *eth_spec;
2879		-
2880		- eth_spec = (struct ib_flow_spec_eth *)flow_spec;
2881		- return is_multicast_ether_addr(eth_spec->mask.dst_mac) &&
2882		- is_multicast_ether_addr(eth_spec->val.dst_mac);
2883		- }
2884		-
2885		- return false;
2886		-}
2887		-
2888		-enum valid_spec {
2889		- VALID_SPEC_INVALID,
2890		- VALID_SPEC_VALID,
2891		- VALID_SPEC_NA,
2892		-};
2893		-
2894		-static enum valid_spec
2895		-is_valid_esp_aes_gcm(struct mlx5_core_dev *mdev,
2896		- const struct mlx5_flow_spec *spec,
2897		- const struct mlx5_flow_act *flow_act,
2898		- bool egress)
2899		-{
2900		- const u32 *match_c = spec->match_criteria;
2901		- bool is_crypto =
2902		- (flow_act->action & (MLX5_FLOW_CONTEXT_ACTION_ENCRYPT \|
2903		- MLX5_FLOW_CONTEXT_ACTION_DECRYPT));
2904		- bool is_ipsec = mlx5_fs_is_ipsec_flow(match_c);
2905		- bool is_drop = flow_act->action & MLX5_FLOW_CONTEXT_ACTION_DROP;
2906		-
2907		- /*
2908		- * Currently only crypto is supported in egress, when regular egress
2909		- * rules would be supported, always return VALID_SPEC_NA.
2910		- */
2911		- if (!is_crypto)
2912		- return egress ? VALID_SPEC_INVALID : VALID_SPEC_NA;
2913		-
2914		- return is_crypto && is_ipsec &&
2915		- (!egress \|\| (!is_drop && !flow_act->has_flow_tag)) ?
2916		- VALID_SPEC_VALID : VALID_SPEC_INVALID;
2917		-}
2918		-
2919		-static bool is_valid_spec(struct mlx5_core_dev *mdev,
2920		- const struct mlx5_flow_spec *spec,
2921		- const struct mlx5_flow_act *flow_act,
2922		- bool egress)
2923		-{
2924		- /* We curretly only support ipsec egress flow */
2925		- return is_valid_esp_aes_gcm(mdev, spec, flow_act, egress) != VALID_SPEC_INVALID;
2926		-}
2927		-
2928		-static bool is_valid_ethertype(struct mlx5_core_dev *mdev,
2929		- const struct ib_flow_attr *flow_attr,
2930		- bool check_inner)
2931		-{
2932		- union ib_flow_spec ib_spec = (union ib_flow_spec )(flow_attr + 1);
2933		- int match_ipv = check_inner ?
2934		- MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
2935		- ft_field_support.inner_ip_version) :
2936		- MLX5_CAP_FLOWTABLE_NIC_RX(mdev,
2937		- ft_field_support.outer_ip_version);
2938		- int inner_bit = check_inner ? IB_FLOW_SPEC_INNER : 0;
2939		- bool ipv4_spec_valid, ipv6_spec_valid;
2940		- unsigned int ip_spec_type = 0;
2941		- bool has_ethertype = false;
2942		- unsigned int spec_index;
2943		- bool mask_valid = true;
2944		- u16 eth_type = 0;
2945		- bool type_valid;
2946		-
2947		- /* Validate that ethertype is correct */
2948		- for (spec_index = 0; spec_index < flow_attr->num_of_specs; spec_index++) {
2949		- if ((ib_spec->type == (IB_FLOW_SPEC_ETH \| inner_bit)) &&
2950		- ib_spec->eth.mask.ether_type) {
2951		- mask_valid = (ib_spec->eth.mask.ether_type ==
2952		- htons(0xffff));
2953		- has_ethertype = true;
2954		- eth_type = ntohs(ib_spec->eth.val.ether_type);
2955		- } else if ((ib_spec->type == (IB_FLOW_SPEC_IPV4 \| inner_bit)) \|\|
2956		- (ib_spec->type == (IB_FLOW_SPEC_IPV6 \| inner_bit))) {
2957		- ip_spec_type = ib_spec->type;
2958		- }
2959		- ib_spec = (void *)ib_spec + ib_spec->size;
2960		- }
2961		-
2962		- type_valid = (!has_ethertype) \|\| (!ip_spec_type);
2963		- if (!type_valid && mask_valid) {
2964		- ipv4_spec_valid = (eth_type == ETH_P_IP) &&
2965		- (ip_spec_type == (IB_FLOW_SPEC_IPV4 \| inner_bit));
2966		- ipv6_spec_valid = (eth_type == ETH_P_IPV6) &&
2967		- (ip_spec_type == (IB_FLOW_SPEC_IPV6 \| inner_bit));
2968		-
2969		- type_valid = (ipv4_spec_valid) \|\| (ipv6_spec_valid) \|\|
2970		- (((eth_type == ETH_P_MPLS_UC) \|\|
2971		- (eth_type == ETH_P_MPLS_MC)) && match_ipv);
2972		- }
2973		-
2974		- return type_valid;
2975		-}
2976		-
2977		-static bool is_valid_attr(struct mlx5_core_dev *mdev,
2978		- const struct ib_flow_attr *flow_attr)
2979		-{
2980		- return is_valid_ethertype(mdev, flow_attr, false) &&
2981		- is_valid_ethertype(mdev, flow_attr, true);
2982		-}
2983		-
2984		-static void put_flow_table(struct mlx5_ib_dev *dev,
2985		- struct mlx5_ib_flow_prio *prio, bool ft_added)
2986		-{
2987		- prio->refcount -= !!ft_added;
2988		- if (!prio->refcount) {
2989		- mlx5_destroy_flow_table(prio->flow_table);
2990		- prio->flow_table = NULL;
2991		- }
2992		-}
2993		-
2994		-static void counters_clear_description(struct ib_counters *counters)
2995		-{
2996		- struct mlx5_ib_mcounters *mcounters = to_mcounters(counters);
2997		-
2998		- mutex_lock(&mcounters->mcntrs_mutex);
2999		- kfree(mcounters->counters_data);
3000		- mcounters->counters_data = NULL;
3001		- mcounters->cntrs_max_index = 0;
3002		- mutex_unlock(&mcounters->mcntrs_mutex);
3003		-}
3004		-
3005		-static int mlx5_ib_destroy_flow(struct ib_flow *flow_id)
3006		-{
3007		- struct mlx5_ib_flow_handler *handler = container_of(flow_id,
3008		- struct mlx5_ib_flow_handler,
3009		- ibflow);
3010		- struct mlx5_ib_flow_handler iter, tmp;
3011		- struct mlx5_ib_dev *dev = handler->dev;
3012		-
3013		- mutex_lock(&dev->flow_db->lock);
3014		-
3015		- list_for_each_entry_safe(iter, tmp, &handler->list, list) {
3016		- mlx5_del_flow_rules(iter->rule);
3017		- put_flow_table(dev, iter->prio, true);
3018		- list_del(&iter->list);
3019		- kfree(iter);
3020		- }
3021		-
3022		- mlx5_del_flow_rules(handler->rule);
3023		- put_flow_table(dev, handler->prio, true);
3024		- if (handler->ibcounters &&
3025		- atomic_read(&handler->ibcounters->usecnt) == 1)
3026		- counters_clear_description(handler->ibcounters);
3027		-
3028		- mutex_unlock(&dev->flow_db->lock);
3029		- if (handler->flow_matcher)
3030		- atomic_dec(&handler->flow_matcher->usecnt);
3031		- kfree(handler);
3032		-
3033		- return 0;
3034		-}
3035		-
3036		-static int ib_prio_to_core_prio(unsigned int priority, bool dont_trap)
3037		-{
3038		- priority *= 2;
3039		- if (!dont_trap)
3040		- priority++;
3041		- return priority;
3042		-}
3043		-
3044		-enum flow_table_type {
3045		- MLX5_IB_FT_RX,
3046		- MLX5_IB_FT_TX
3047		-};
3048		-
3049		-#define MLX5_FS_MAX_TYPES 6
3050		-#define MLX5_FS_MAX_ENTRIES BIT(16)
3051		-
3052		-static struct mlx5_ib_flow_prio _get_prio(struct mlx5_flow_namespace ns,
3053		- struct mlx5_ib_flow_prio *prio,
3054		- int priority,
3055		- int num_entries, int num_groups)
3056		-{
3057		- struct mlx5_flow_table *ft;
3058		-
3059		- ft = mlx5_create_auto_grouped_flow_table(ns, priority,
3060		- num_entries,
3061		- num_groups,
3062		- 0, 0);
3063		- if (IS_ERR(ft))
3064		- return ERR_CAST(ft);
3065		-
3066		- prio->flow_table = ft;
3067		- prio->refcount = 0;
3068		- return prio;
3069		-}
3070		-
3071		-static struct mlx5_ib_flow_prio get_flow_table(struct mlx5_ib_dev dev,
3072		- struct ib_flow_attr *flow_attr,
3073		- enum flow_table_type ft_type)
3074		-{
3075		- bool dont_trap = flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP;
3076		- struct mlx5_flow_namespace *ns = NULL;
3077		- struct mlx5_ib_flow_prio *prio;
3078		- struct mlx5_flow_table *ft;
3079		- int max_table_size;
3080		- int num_entries;
3081		- int num_groups;
3082		- int priority;
3083		-
3084		- max_table_size = BIT(MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
3085		- log_max_ft_size));
3086		- if (flow_attr->type == IB_FLOW_ATTR_NORMAL) {
3087		- if (ft_type == MLX5_IB_FT_TX)
3088		- priority = 0;
3089		- else if (flow_is_multicast_only(flow_attr) &&
3090		- !dont_trap)
3091		- priority = MLX5_IB_FLOW_MCAST_PRIO;
3092		- else
3093		- priority = ib_prio_to_core_prio(flow_attr->priority,
3094		- dont_trap);
3095		- ns = mlx5_get_flow_namespace(dev->mdev,
3096		- ft_type == MLX5_IB_FT_TX ?
3097		- MLX5_FLOW_NAMESPACE_EGRESS :
3098		- MLX5_FLOW_NAMESPACE_BYPASS);
3099		- num_entries = MLX5_FS_MAX_ENTRIES;
3100		- num_groups = MLX5_FS_MAX_TYPES;
3101		- prio = &dev->flow_db->prios[priority];
3102		- } else if (flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT \|\|
3103		- flow_attr->type == IB_FLOW_ATTR_MC_DEFAULT) {
3104		- ns = mlx5_get_flow_namespace(dev->mdev,
3105		- MLX5_FLOW_NAMESPACE_LEFTOVERS);
3106		- build_leftovers_ft_param(&priority,
3107		- &num_entries,
3108		- &num_groups);
3109		- prio = &dev->flow_db->prios[MLX5_IB_FLOW_LEFTOVERS_PRIO];
3110		- } else if (flow_attr->type == IB_FLOW_ATTR_SNIFFER) {
3111		- if (!MLX5_CAP_FLOWTABLE(dev->mdev,
3112		- allow_sniffer_and_nic_rx_shared_tir))
3113		- return ERR_PTR(-ENOTSUPP);
3114		-
3115		- ns = mlx5_get_flow_namespace(dev->mdev, ft_type == MLX5_IB_FT_RX ?
3116		- MLX5_FLOW_NAMESPACE_SNIFFER_RX :
3117		- MLX5_FLOW_NAMESPACE_SNIFFER_TX);
3118		-
3119		- prio = &dev->flow_db->sniffer[ft_type];
3120		- priority = 0;
3121		- num_entries = 1;
3122		- num_groups = 1;
3123		- }
3124		-
3125		- if (!ns)
3126		- return ERR_PTR(-ENOTSUPP);
3127		-
3128		- if (num_entries > max_table_size)
3129		- return ERR_PTR(-ENOMEM);
3130		-
3131		- ft = prio->flow_table;
3132		- if (!ft)
3133		- return _get_prio(ns, prio, priority, num_entries, num_groups);
3134		-
3135		- return prio;
3136		-}
3137		-
3138		-static void set_underlay_qp(struct mlx5_ib_dev *dev,
3139		- struct mlx5_flow_spec *spec,
3140		- u32 underlay_qpn)
3141		-{
3142		- void *misc_params_c = MLX5_ADDR_OF(fte_match_param,
3143		- spec->match_criteria,
3144		- misc_parameters);
3145		- void *misc_params_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
3146		- misc_parameters);
3147		-
3148		- if (underlay_qpn &&
3149		- MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
3150		- ft_field_support.bth_dst_qp)) {
3151		- MLX5_SET(fte_match_set_misc,
3152		- misc_params_v, bth_dst_qp, underlay_qpn);
3153		- MLX5_SET(fte_match_set_misc,
3154		- misc_params_c, bth_dst_qp, 0xffffff);
3155		- }
3156		-}
3157		-
3158		-static int read_flow_counters(struct ib_device *ibdev,
3159		- struct mlx5_read_counters_attr *read_attr)
3160		-{
3161		- struct mlx5_fc *fc = read_attr->hw_cntrs_hndl;
3162		- struct mlx5_ib_dev *dev = to_mdev(ibdev);
3163		-
3164		- return mlx5_fc_query(dev->mdev, fc,
3165		- &read_attr->out[IB_COUNTER_PACKETS],
3166		- &read_attr->out[IB_COUNTER_BYTES]);
3167		-}
3168		-
3169		-/* flow counters currently expose two counters packets and bytes */
3170		-#define FLOW_COUNTERS_NUM 2
3171		-static int counters_set_description(struct ib_counters *counters,
3172		- enum mlx5_ib_counters_type counters_type,
3173		- struct mlx5_ib_flow_counters_desc *desc_data,
3174		- u32 ncounters)
3175		-{
3176		- struct mlx5_ib_mcounters *mcounters = to_mcounters(counters);
3177		- u32 cntrs_max_index = 0;
3178		- int i;
3179		-
3180		- if (counters_type != MLX5_IB_COUNTERS_FLOW)
3181		- return -EINVAL;
3182		-
3183		- /* init the fields for the object */
3184		- mcounters->type = counters_type;
3185		- mcounters->read_counters = read_flow_counters;
3186		- mcounters->counters_num = FLOW_COUNTERS_NUM;
3187		- mcounters->ncounters = ncounters;
3188		- /* each counter entry have both description and index pair */
3189		- for (i = 0; i < ncounters; i++) {
3190		- if (desc_data[i].description > IB_COUNTER_BYTES)
3191		- return -EINVAL;
3192		-
3193		- if (cntrs_max_index <= desc_data[i].index)
3194		- cntrs_max_index = desc_data[i].index + 1;
3195		- }
3196		-
3197		- mutex_lock(&mcounters->mcntrs_mutex);
3198		- mcounters->counters_data = desc_data;
3199		- mcounters->cntrs_max_index = cntrs_max_index;
3200		- mutex_unlock(&mcounters->mcntrs_mutex);
3201		-
3202		- return 0;
3203		-}
3204		-
3205		-#define MAX_COUNTERS_NUM (USHRT_MAX / (sizeof(u32) * 2))
3206		-static int flow_counters_set_data(struct ib_counters *ibcounters,
3207		- struct mlx5_ib_create_flow *ucmd)
3208		-{
3209		- struct mlx5_ib_mcounters *mcounters = to_mcounters(ibcounters);
3210		- struct mlx5_ib_flow_counters_data *cntrs_data = NULL;
3211		- struct mlx5_ib_flow_counters_desc *desc_data = NULL;
3212		- bool hw_hndl = false;
3213		- int ret = 0;
3214		-
3215		- if (ucmd && ucmd->ncounters_data != 0) {
3216		- cntrs_data = ucmd->data;
3217		- if (cntrs_data->ncounters > MAX_COUNTERS_NUM)
3218		- return -EINVAL;
3219		-
3220		- desc_data = kcalloc(cntrs_data->ncounters,
3221		- sizeof(*desc_data),
3222		- GFP_KERNEL);
3223		- if (!desc_data)
3224		- return -ENOMEM;
3225		-
3226		- if (copy_from_user(desc_data,
3227		- u64_to_user_ptr(cntrs_data->counters_data),
3228		- sizeof(desc_data) cntrs_data->ncounters)) {
3229		- ret = -EFAULT;
3230		- goto free;
3231		- }
3232		- }
3233		-
3234		- if (!mcounters->hw_cntrs_hndl) {
3235		- mcounters->hw_cntrs_hndl = mlx5_fc_create(
3236		- to_mdev(ibcounters->device)->mdev, false);
3237		- if (IS_ERR(mcounters->hw_cntrs_hndl)) {
3238		- ret = PTR_ERR(mcounters->hw_cntrs_hndl);
3239		- goto free;
3240		- }
3241		- hw_hndl = true;
3242		- }
3243		-
3244		- if (desc_data) {
3245		- /* counters already bound to at least one flow */
3246		- if (mcounters->cntrs_max_index) {
3247		- ret = -EINVAL;
3248		- goto free_hndl;
3249		- }
3250		-
3251		- ret = counters_set_description(ibcounters,
3252		- MLX5_IB_COUNTERS_FLOW,
3253		- desc_data,
3254		- cntrs_data->ncounters);
3255		- if (ret)
3256		- goto free_hndl;
3257		-
3258		- } else if (!mcounters->cntrs_max_index) {
3259		- /* counters not bound yet, must have udata passed */
3260		- ret = -EINVAL;
3261		- goto free_hndl;
3262		- }
3263		-
3264		- return 0;
3265		-
3266		-free_hndl:
3267		- if (hw_hndl) {
3268		- mlx5_fc_destroy(to_mdev(ibcounters->device)->mdev,
3269		- mcounters->hw_cntrs_hndl);
3270		- mcounters->hw_cntrs_hndl = NULL;
3271		- }
3272		-free:
3273		- kfree(desc_data);
3274		- return ret;
3275		-}
3276		-
3277		-static struct mlx5_ib_flow_handler _create_flow_rule(struct mlx5_ib_dev dev,
3278		- struct mlx5_ib_flow_prio *ft_prio,
3279		- const struct ib_flow_attr *flow_attr,
3280		- struct mlx5_flow_destination *dst,
3281		- u32 underlay_qpn,
3282		- struct mlx5_ib_create_flow *ucmd)
3283		-{
3284		- struct mlx5_flow_table *ft = ft_prio->flow_table;
3285		- struct mlx5_ib_flow_handler *handler;
3286		- struct mlx5_flow_act flow_act = {.flow_tag = MLX5_FS_DEFAULT_FLOW_TAG};
3287		- struct mlx5_flow_spec *spec;
3288		- struct mlx5_flow_destination dest_arr[2] = {};
3289		- struct mlx5_flow_destination *rule_dst = dest_arr;
3290		- const void ib_flow = (const void )flow_attr + sizeof(*flow_attr);
3291		- unsigned int spec_index;
3292		- u32 prev_type = 0;
3293		- int err = 0;
3294		- int dest_num = 0;
3295		- bool is_egress = flow_attr->flags & IB_FLOW_ATTR_FLAGS_EGRESS;
3296		-
3297		- if (!is_valid_attr(dev->mdev, flow_attr))
3298		- return ERR_PTR(-EINVAL);
3299		-
3300		- spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
3301		- handler = kzalloc(sizeof(*handler), GFP_KERNEL);
3302		- if (!handler \|\| !spec) {
3303		- err = -ENOMEM;
3304		- goto free;
3305		- }
3306		-
3307		- INIT_LIST_HEAD(&handler->list);
3308		-
3309		- for (spec_index = 0; spec_index < flow_attr->num_of_specs; spec_index++) {
3310		- err = parse_flow_attr(dev->mdev, spec->match_criteria,
3311		- spec->match_value,
3312		- ib_flow, flow_attr, &flow_act,
3313		- prev_type);
3314		- if (err < 0)
3315		- goto free;
3316		-
3317		- prev_type = ((union ib_flow_spec *)ib_flow)->type;
3318		- ib_flow += ((union ib_flow_spec *)ib_flow)->size;
3319		- }
3320		-
3321		- if (dst && !(flow_act.action & MLX5_FLOW_CONTEXT_ACTION_DROP)) {
3322		- memcpy(&dest_arr[0], dst, sizeof(*dst));
3323		- dest_num++;
3324		- }
3325		-
3326		- if (!flow_is_multicast_only(flow_attr))
3327		- set_underlay_qp(dev, spec, underlay_qpn);
3328		-
3329		- if (dev->rep) {
3330		- void *misc;
3331		-
3332		- misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
3333		- misc_parameters);
3334		- MLX5_SET(fte_match_set_misc, misc, source_port,
3335		- dev->rep->vport);
3336		- misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
3337		- misc_parameters);
3338		- MLX5_SET_TO_ONES(fte_match_set_misc, misc, source_port);
3339		- }
3340		-
3341		- spec->match_criteria_enable = get_match_criteria_enable(spec->match_criteria);
3342		-
3343		- if (is_egress &&
3344		- !is_valid_spec(dev->mdev, spec, &flow_act, is_egress)) {
3345		- err = -EINVAL;
3346		- goto free;
3347		- }
3348		-
3349		- if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
3350		- err = flow_counters_set_data(flow_act.counters, ucmd);
3351		- if (err)
3352		- goto free;
3353		-
3354		- handler->ibcounters = flow_act.counters;
3355		- dest_arr[dest_num].type =
3356		- MLX5_FLOW_DESTINATION_TYPE_COUNTER;
3357		- dest_arr[dest_num].counter =
3358		- to_mcounters(flow_act.counters)->hw_cntrs_hndl;
3359		- dest_num++;
3360		- }
3361		-
3362		- if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_DROP) {
3363		- if (!dest_num)
3364		- rule_dst = NULL;
3365		- } else {
3366		- if (is_egress)
3367		- flow_act.action \|= MLX5_FLOW_CONTEXT_ACTION_ALLOW;
3368		- else
3369		- flow_act.action \|=
3370		- dest_num ? MLX5_FLOW_CONTEXT_ACTION_FWD_DEST :
3371		- MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
3372		- }
3373		-
3374		- if (flow_act.has_flow_tag &&
3375		- (flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT \|\|
3376		- flow_attr->type == IB_FLOW_ATTR_MC_DEFAULT)) {
3377		- mlx5_ib_warn(dev, "Flow tag %u and attribute type %x isn't allowed in leftovers\n",
3378		- flow_act.flow_tag, flow_attr->type);
3379		- err = -EINVAL;
3380		- goto free;
3381		- }
3382		- handler->rule = mlx5_add_flow_rules(ft, spec,
3383		- &flow_act,
3384		- rule_dst, dest_num);
3385		-
3386		- if (IS_ERR(handler->rule)) {
3387		- err = PTR_ERR(handler->rule);
3388		- goto free;
3389		- }
3390		-
3391		- ft_prio->refcount++;
3392		- handler->prio = ft_prio;
3393		- handler->dev = dev;
3394		-
3395		- ft_prio->flow_table = ft;
3396		-free:
3397		- if (err && handler) {
3398		- if (handler->ibcounters &&
3399		- atomic_read(&handler->ibcounters->usecnt) == 1)
3400		- counters_clear_description(handler->ibcounters);
3401		- kfree(handler);
3402		- }
3403		- kvfree(spec);
3404		- return err ? ERR_PTR(err) : handler;
3405		-}
3406		-
3407		-static struct mlx5_ib_flow_handler create_flow_rule(struct mlx5_ib_dev dev,
3408		- struct mlx5_ib_flow_prio *ft_prio,
3409		- const struct ib_flow_attr *flow_attr,
3410		- struct mlx5_flow_destination *dst)
3411		-{
3412		- return _create_flow_rule(dev, ft_prio, flow_attr, dst, 0, NULL);
3413		-}
3414		-
3415		-static struct mlx5_ib_flow_handler create_dont_trap_rule(struct mlx5_ib_dev dev,
3416		- struct mlx5_ib_flow_prio *ft_prio,
3417		- struct ib_flow_attr *flow_attr,
3418		- struct mlx5_flow_destination *dst)
3419		-{
3420		- struct mlx5_ib_flow_handler *handler_dst = NULL;
3421		- struct mlx5_ib_flow_handler *handler = NULL;
3422		-
3423		- handler = create_flow_rule(dev, ft_prio, flow_attr, NULL);
3424		- if (!IS_ERR(handler)) {
3425		- handler_dst = create_flow_rule(dev, ft_prio,
3426		- flow_attr, dst);
3427		- if (IS_ERR(handler_dst)) {
3428		- mlx5_del_flow_rules(handler->rule);
3429		- ft_prio->refcount--;
3430		- kfree(handler);
3431		- handler = handler_dst;
3432		- } else {
3433		- list_add(&handler_dst->list, &handler->list);
3434		- }
3435		- }
3436		-
3437		- return handler;
3438		-}
3439		-enum {
3440		- LEFTOVERS_MC,
3441		- LEFTOVERS_UC,
3442		-};
3443		-
3444		-static struct mlx5_ib_flow_handler create_leftovers_rule(struct mlx5_ib_dev dev,
3445		- struct mlx5_ib_flow_prio *ft_prio,
3446		- struct ib_flow_attr *flow_attr,
3447		- struct mlx5_flow_destination *dst)
3448		-{
3449		- struct mlx5_ib_flow_handler *handler_ucast = NULL;
3450		- struct mlx5_ib_flow_handler *handler = NULL;
3451		-
3452		- static struct {
3453		- struct ib_flow_attr flow_attr;
3454		- struct ib_flow_spec_eth eth_flow;
3455		- } leftovers_specs[] = {
3456		- [LEFTOVERS_MC] = {
3457		- .flow_attr = {
3458		- .num_of_specs = 1,
3459		- .size = sizeof(leftovers_specs[0])
3460		- },
3461		- .eth_flow = {
3462		- .type = IB_FLOW_SPEC_ETH,
3463		- .size = sizeof(struct ib_flow_spec_eth),
3464		- .mask = {.dst_mac = {0x1} },
3465		- .val = {.dst_mac = {0x1} }
3466		- }
3467		- },
3468		- [LEFTOVERS_UC] = {
3469		- .flow_attr = {
3470		- .num_of_specs = 1,
3471		- .size = sizeof(leftovers_specs[0])
3472		- },
3473		- .eth_flow = {
3474		- .type = IB_FLOW_SPEC_ETH,
3475		- .size = sizeof(struct ib_flow_spec_eth),
3476		- .mask = {.dst_mac = {0x1} },
3477		- .val = {.dst_mac = {} }
3478		- }
3479		- }
3480		- };
3481		-
3482		- handler = create_flow_rule(dev, ft_prio,
3483		- &leftovers_specs[LEFTOVERS_MC].flow_attr,
3484		- dst);
3485		- if (!IS_ERR(handler) &&
3486		- flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT) {
3487		- handler_ucast = create_flow_rule(dev, ft_prio,
3488		- &leftovers_specs[LEFTOVERS_UC].flow_attr,
3489		- dst);
3490		- if (IS_ERR(handler_ucast)) {
3491		- mlx5_del_flow_rules(handler->rule);
3492		- ft_prio->refcount--;
3493		- kfree(handler);
3494		- handler = handler_ucast;
3495		- } else {
3496		- list_add(&handler_ucast->list, &handler->list);
3497		- }
3498		- }
3499		-
3500		- return handler;
3501		-}
3502		-
3503		-static struct mlx5_ib_flow_handler create_sniffer_rule(struct mlx5_ib_dev dev,
3504		- struct mlx5_ib_flow_prio *ft_rx,
3505		- struct mlx5_ib_flow_prio *ft_tx,
3506		- struct mlx5_flow_destination *dst)
3507		-{
3508		- struct mlx5_ib_flow_handler *handler_rx;
3509		- struct mlx5_ib_flow_handler *handler_tx;
3510		- int err;
3511		- static const struct ib_flow_attr flow_attr = {
3512		- .num_of_specs = 0,
3513		- .size = sizeof(flow_attr)
3514		- };
3515		-
3516		- handler_rx = create_flow_rule(dev, ft_rx, &flow_attr, dst);
3517		- if (IS_ERR(handler_rx)) {
3518		- err = PTR_ERR(handler_rx);
3519		- goto err;
3520		- }
3521		-
3522		- handler_tx = create_flow_rule(dev, ft_tx, &flow_attr, dst);
3523		- if (IS_ERR(handler_tx)) {
3524		- err = PTR_ERR(handler_tx);
3525		- goto err_tx;
3526		- }
3527		-
3528		- list_add(&handler_tx->list, &handler_rx->list);
3529		-
3530		- return handler_rx;
3531		-
3532		-err_tx:
3533		- mlx5_del_flow_rules(handler_rx->rule);
3534		- ft_rx->refcount--;
3535		- kfree(handler_rx);
3536		-err:
3537		- return ERR_PTR(err);
3538		-}
3539		-
3540		-static struct ib_flow mlx5_ib_create_flow(struct ib_qp qp,
3541		- struct ib_flow_attr *flow_attr,
3542		- int domain,
3543		- struct ib_udata *udata)
3544		-{
3545		- struct mlx5_ib_dev *dev = to_mdev(qp->device);
3546		- struct mlx5_ib_qp *mqp = to_mqp(qp);
3547		- struct mlx5_ib_flow_handler *handler = NULL;
3548		- struct mlx5_flow_destination *dst = NULL;
3549		- struct mlx5_ib_flow_prio *ft_prio_tx = NULL;
3550		- struct mlx5_ib_flow_prio *ft_prio;
3551		- bool is_egress = flow_attr->flags & IB_FLOW_ATTR_FLAGS_EGRESS;
3552		- struct mlx5_ib_create_flow *ucmd = NULL, ucmd_hdr;
3553		- size_t min_ucmd_sz, required_ucmd_sz;
3554		- int err;
3555		- int underlay_qpn;
3556		-
3557		- if (udata && udata->inlen) {
3558		- min_ucmd_sz = offsetof(typeof(ucmd_hdr), reserved) +
3559		- sizeof(ucmd_hdr.reserved);
3560		- if (udata->inlen < min_ucmd_sz)
3561		- return ERR_PTR(-EOPNOTSUPP);
3562		-
3563		- err = ib_copy_from_udata(&ucmd_hdr, udata, min_ucmd_sz);
3564		- if (err)
3565		- return ERR_PTR(err);
3566		-
3567		- /* currently supports only one counters data */
3568		- if (ucmd_hdr.ncounters_data > 1)
3569		- return ERR_PTR(-EINVAL);
3570		-
3571		- required_ucmd_sz = min_ucmd_sz +
3572		- sizeof(struct mlx5_ib_flow_counters_data) *
3573		- ucmd_hdr.ncounters_data;
3574		- if (udata->inlen > required_ucmd_sz &&
3575		- !ib_is_udata_cleared(udata, required_ucmd_sz,
3576		- udata->inlen - required_ucmd_sz))
3577		- return ERR_PTR(-EOPNOTSUPP);
3578		-
3579		- ucmd = kzalloc(required_ucmd_sz, GFP_KERNEL);
3580		- if (!ucmd)
3581		- return ERR_PTR(-ENOMEM);
3582		-
3583		- err = ib_copy_from_udata(ucmd, udata, required_ucmd_sz);
3584		- if (err)
3585		- goto free_ucmd;
3586		- }
3587		-
3588		- if (flow_attr->priority > MLX5_IB_FLOW_LAST_PRIO) {
3589		- err = -ENOMEM;
3590		- goto free_ucmd;
3591		- }
3592		-
3593		- if (domain != IB_FLOW_DOMAIN_USER \|\|
3594		- flow_attr->port > dev->num_ports \|\|
3595		- (flow_attr->flags & ~(IB_FLOW_ATTR_FLAGS_DONT_TRAP \|
3596		- IB_FLOW_ATTR_FLAGS_EGRESS))) {
3597		- err = -EINVAL;
3598		- goto free_ucmd;
3599		- }
3600		-
3601		- if (is_egress &&
3602		- (flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT \|\|
3603		- flow_attr->type == IB_FLOW_ATTR_MC_DEFAULT)) {
3604		- err = -EINVAL;
3605		- goto free_ucmd;
3606		- }
3607		-
3608		- dst = kzalloc(sizeof(*dst), GFP_KERNEL);
3609		- if (!dst) {
3610		- err = -ENOMEM;
3611		- goto free_ucmd;
3612		- }
3613		-
3614		- mutex_lock(&dev->flow_db->lock);
3615		-
3616		- ft_prio = get_flow_table(dev, flow_attr,
3617		- is_egress ? MLX5_IB_FT_TX : MLX5_IB_FT_RX);
3618		- if (IS_ERR(ft_prio)) {
3619		- err = PTR_ERR(ft_prio);
3620		- goto unlock;
3621		- }
3622		- if (flow_attr->type == IB_FLOW_ATTR_SNIFFER) {
3623		- ft_prio_tx = get_flow_table(dev, flow_attr, MLX5_IB_FT_TX);
3624		- if (IS_ERR(ft_prio_tx)) {
3625		- err = PTR_ERR(ft_prio_tx);
3626		- ft_prio_tx = NULL;
3627		- goto destroy_ft;
3628		- }
3629		- }
3630		-
3631		- if (is_egress) {
3632		- dst->type = MLX5_FLOW_DESTINATION_TYPE_PORT;
3633		- } else {
3634		- dst->type = MLX5_FLOW_DESTINATION_TYPE_TIR;
3635		- if (mqp->flags & MLX5_IB_QP_RSS)
3636		- dst->tir_num = mqp->rss_qp.tirn;
3637		- else
3638		- dst->tir_num = mqp->raw_packet_qp.rq.tirn;
3639		- }
3640		-
3641		- if (flow_attr->type == IB_FLOW_ATTR_NORMAL) {
3642		- if (flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) {
3643		- handler = create_dont_trap_rule(dev, ft_prio,
3644		- flow_attr, dst);
3645		- } else {
3646		- underlay_qpn = (mqp->flags & MLX5_IB_QP_UNDERLAY) ?
3647		- mqp->underlay_qpn : 0;
3648		- handler = _create_flow_rule(dev, ft_prio, flow_attr,
3649		- dst, underlay_qpn, ucmd);
3650		- }
3651		- } else if (flow_attr->type == IB_FLOW_ATTR_ALL_DEFAULT \|\|
3652		- flow_attr->type == IB_FLOW_ATTR_MC_DEFAULT) {
3653		- handler = create_leftovers_rule(dev, ft_prio, flow_attr,
3654		- dst);
3655		- } else if (flow_attr->type == IB_FLOW_ATTR_SNIFFER) {
3656		- handler = create_sniffer_rule(dev, ft_prio, ft_prio_tx, dst);
3657		- } else {
3658		- err = -EINVAL;
3659		- goto destroy_ft;
3660		- }
3661		-
3662		- if (IS_ERR(handler)) {
3663		- err = PTR_ERR(handler);
3664		- handler = NULL;
3665		- goto destroy_ft;
3666		- }
3667		-
3668		- mutex_unlock(&dev->flow_db->lock);
3669		- kfree(dst);
3670		- kfree(ucmd);
3671		-
3672		- return &handler->ibflow;
3673		-
3674		-destroy_ft:
3675		- put_flow_table(dev, ft_prio, false);
3676		- if (ft_prio_tx)
3677		- put_flow_table(dev, ft_prio_tx, false);
3678		-unlock:
3679		- mutex_unlock(&dev->flow_db->lock);
3680		- kfree(dst);
3681		-free_ucmd:
3682		- kfree(ucmd);
3683		- return ERR_PTR(err);
3684		-}
3685		-
3686		-static struct mlx5_ib_flow_prio _get_flow_table(struct mlx5_ib_dev dev,
3687		- int priority, bool mcast)
3688		-{
3689		- int max_table_size;
3690		- struct mlx5_flow_namespace *ns = NULL;
3691		- struct mlx5_ib_flow_prio *prio;
3692		-
3693		- max_table_size = BIT(MLX5_CAP_FLOWTABLE_NIC_RX(dev->mdev,
3694		- log_max_ft_size));
3695		- if (max_table_size < MLX5_FS_MAX_ENTRIES)
3696		- return ERR_PTR(-ENOMEM);
3697		-
3698		- if (mcast)
3699		- priority = MLX5_IB_FLOW_MCAST_PRIO;
3700		- else
3701		- priority = ib_prio_to_core_prio(priority, false);
3702		-
3703		- ns = mlx5_get_flow_namespace(dev->mdev, MLX5_FLOW_NAMESPACE_BYPASS);
3704		- if (!ns)
3705		- return ERR_PTR(-ENOTSUPP);
3706		-
3707		- prio = &dev->flow_db->prios[priority];
3708		-
3709		- if (prio->flow_table)
3710		- return prio;
3711		-
3712		- return _get_prio(ns, prio, priority, MLX5_FS_MAX_ENTRIES,
3713		- MLX5_FS_MAX_TYPES);
3714		-}
3715		-
3716		-static struct mlx5_ib_flow_handler *
3717		-_create_raw_flow_rule(struct mlx5_ib_dev *dev,
3718		- struct mlx5_ib_flow_prio *ft_prio,
3719		- struct mlx5_flow_destination *dst,
3720		- struct mlx5_ib_flow_matcher *fs_matcher,
3721		- void *cmd_in, int inlen)
3722		-{
3723		- struct mlx5_ib_flow_handler *handler;
3724		- struct mlx5_flow_act flow_act = {.flow_tag = MLX5_FS_DEFAULT_FLOW_TAG};
3725		- struct mlx5_flow_spec *spec;
3726		- struct mlx5_flow_table *ft = ft_prio->flow_table;
3727		- int err = 0;
3728		-
3729		- spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
3730		- handler = kzalloc(sizeof(*handler), GFP_KERNEL);
3731		- if (!handler \|\| !spec) {
3732		- err = -ENOMEM;
3733		- goto free;
3734		- }
3735		-
3736		- INIT_LIST_HEAD(&handler->list);
3737		-
3738		- memcpy(spec->match_value, cmd_in, inlen);
3739		- memcpy(spec->match_criteria, fs_matcher->matcher_mask.match_params,
3740		- fs_matcher->mask_len);
3741		- spec->match_criteria_enable = fs_matcher->match_criteria_enable;
3742		-
3743		- flow_act.action \|= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
3744		- handler->rule = mlx5_add_flow_rules(ft, spec,
3745		- &flow_act, dst, 1);
3746		-
3747		- if (IS_ERR(handler->rule)) {
3748		- err = PTR_ERR(handler->rule);
3749		- goto free;
3750		- }
3751		-
3752		- ft_prio->refcount++;
3753		- handler->prio = ft_prio;
3754		- handler->dev = dev;
3755		- ft_prio->flow_table = ft;
3756		-
3757		-free:
3758		- if (err)
3759		- kfree(handler);
3760		- kvfree(spec);
3761		- return err ? ERR_PTR(err) : handler;
3762		-}
3763		-
3764		-static bool raw_fs_is_multicast(struct mlx5_ib_flow_matcher *fs_matcher,
3765		- void *match_v)
3766		-{
3767		- void *match_c;
3768		- void match_v_set_lyr_2_4, match_c_set_lyr_2_4;
3769		- void dmac, dmac_mask;
3770		- void ipv4, ipv4_mask;
3771		-
3772		- if (!(fs_matcher->match_criteria_enable &
3773		- (1 << MATCH_CRITERIA_ENABLE_OUTER_BIT)))
3774		- return false;
3775		-
3776		- match_c = fs_matcher->matcher_mask.match_params;
3777		- match_v_set_lyr_2_4 = MLX5_ADDR_OF(fte_match_param, match_v,
3778		- outer_headers);
3779		- match_c_set_lyr_2_4 = MLX5_ADDR_OF(fte_match_param, match_c,
3780		- outer_headers);
3781		-
3782		- dmac = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_v_set_lyr_2_4,
3783		- dmac_47_16);
3784		- dmac_mask = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_c_set_lyr_2_4,
3785		- dmac_47_16);
3786		-
3787		- if (is_multicast_ether_addr(dmac) &&
3788		- is_multicast_ether_addr(dmac_mask))
3789		- return true;
3790		-
3791		- ipv4 = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_v_set_lyr_2_4,
3792		- dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
3793		-
3794		- ipv4_mask = MLX5_ADDR_OF(fte_match_set_lyr_2_4, match_c_set_lyr_2_4,
3795		- dst_ipv4_dst_ipv6.ipv4_layout.ipv4);
3796		-
3797		- if (ipv4_is_multicast((__be32 )(ipv4)) &&
3798		- ipv4_is_multicast((__be32 )(ipv4_mask)))
3799		- return true;
3800		-
3801		- return false;
3802		-}
3803		-
3804		-struct mlx5_ib_flow_handler *
3805		-mlx5_ib_raw_fs_rule_add(struct mlx5_ib_dev *dev,
3806		- struct mlx5_ib_flow_matcher *fs_matcher,
3807		- void *cmd_in, int inlen, int dest_id,
3808		- int dest_type)
3809		-{
3810		- struct mlx5_flow_destination *dst;
3811		- struct mlx5_ib_flow_prio *ft_prio;
3812		- int priority = fs_matcher->priority;
3813		- struct mlx5_ib_flow_handler *handler;
3814		- bool mcast;
3815		- int err;
3816		-
3817		- if (fs_matcher->flow_type != MLX5_IB_FLOW_TYPE_NORMAL)
3818		- return ERR_PTR(-EOPNOTSUPP);
3819		-
3820		- if (fs_matcher->priority > MLX5_IB_FLOW_LAST_PRIO)
3821		- return ERR_PTR(-ENOMEM);
3822		-
3823		- dst = kzalloc(sizeof(*dst), GFP_KERNEL);
3824		- if (!dst)
3825		- return ERR_PTR(-ENOMEM);
3826		-
3827		- mcast = raw_fs_is_multicast(fs_matcher, cmd_in);
3828		- mutex_lock(&dev->flow_db->lock);
3829		-
3830		- ft_prio = _get_flow_table(dev, priority, mcast);
3831		- if (IS_ERR(ft_prio)) {
3832		- err = PTR_ERR(ft_prio);
3833		- goto unlock;
3834		- }
3835		-
3836		- if (dest_type == MLX5_FLOW_DESTINATION_TYPE_TIR) {
3837		- dst->type = dest_type;
3838		- dst->tir_num = dest_id;
3839		- } else {
3840		- dst->type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM;
3841		- dst->ft_num = dest_id;
3842		- }
3843		-
3844		- handler = _create_raw_flow_rule(dev, ft_prio, dst, fs_matcher, cmd_in,
3845		- inlen);
3846		-
3847		- if (IS_ERR(handler)) {
3848		- err = PTR_ERR(handler);
3849		- goto destroy_ft;
3850		- }
3851		-
3852		- mutex_unlock(&dev->flow_db->lock);
3853		- atomic_inc(&fs_matcher->usecnt);
3854		- handler->flow_matcher = fs_matcher;
3855		-
3856		- kfree(dst);
3857		-
3858		- return handler;
3859		-
3860		-destroy_ft:
3861		- put_flow_table(dev, ft_prio, false);
3862		-unlock:
3863		- mutex_unlock(&dev->flow_db->lock);
3864		- kfree(dst);
3865		-
3866		- return ERR_PTR(err);
3867		-}
3868		-
3869		-static u32 mlx5_ib_flow_action_flags_to_accel_xfrm_flags(u32 mlx5_flags)
3870		-{
3871		- u32 flags = 0;
3872		-
3873		- if (mlx5_flags & MLX5_IB_UAPI_FLOW_ACTION_FLAGS_REQUIRE_METADATA)
3874		- flags \|= MLX5_ACCEL_XFRM_FLAG_REQUIRE_METADATA;
3875		-
3876		- return flags;
3877		-}
3878		-
3879		-#define MLX5_FLOW_ACTION_ESP_CREATE_LAST_SUPPORTED MLX5_IB_UAPI_FLOW_ACTION_FLAGS_REQUIRE_METADATA
3880		-static struct ib_flow_action *
3881		-mlx5_ib_create_flow_action_esp(struct ib_device *device,
3882		- const struct ib_flow_action_attrs_esp *attr,
3883		- struct uverbs_attr_bundle *attrs)
3884		-{
3885		- struct mlx5_ib_dev *mdev = to_mdev(device);
3886		- struct ib_uverbs_flow_action_esp_keymat_aes_gcm *aes_gcm;
3887		- struct mlx5_accel_esp_xfrm_attrs accel_attrs = {};
3888		- struct mlx5_ib_flow_action *action;
3889		- u64 action_flags;
3890		- u64 flags;
3891		- int err = 0;
3892		-
3893		- err = uverbs_get_flags64(
3894		- &action_flags, attrs, MLX5_IB_ATTR_CREATE_FLOW_ACTION_FLAGS,
3895		- ((MLX5_FLOW_ACTION_ESP_CREATE_LAST_SUPPORTED << 1) - 1));
3896		- if (err)
3897		- return ERR_PTR(err);
3898		-
3899		- flags = mlx5_ib_flow_action_flags_to_accel_xfrm_flags(action_flags);
3900		-
3901		- /* We current only support a subset of the standard features. Only a
3902		- * keymat of type AES_GCM, with icv_len == 16, iv_algo == SEQ and esn
3903		- * (with overlap). Full offload mode isn't supported.
3904		- */
3905		- if (!attr->keymat \|\| attr->replay \|\| attr->encap \|\|
3906		- attr->spi \|\| attr->seq \|\| attr->tfc_pad \|\|
3907		- attr->hard_limit_pkts \|\|
3908		- (attr->flags & ~(IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED \|
3909		- IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ENCRYPT)))
3910		- return ERR_PTR(-EOPNOTSUPP);
3911		-
3912		- if (attr->keymat->protocol !=
3913		- IB_UVERBS_FLOW_ACTION_ESP_KEYMAT_AES_GCM)
3914		- return ERR_PTR(-EOPNOTSUPP);
3915		-
3916		- aes_gcm = &attr->keymat->keymat.aes_gcm;
3917		-
3918		- if (aes_gcm->icv_len != 16 \|\|
3919		- aes_gcm->iv_algo != IB_UVERBS_FLOW_ACTION_IV_ALGO_SEQ)
3920		- return ERR_PTR(-EOPNOTSUPP);
3921		-
3922		- action = kmalloc(sizeof(*action), GFP_KERNEL);
3923		- if (!action)
3924		- return ERR_PTR(-ENOMEM);
3925		-
3926		- action->esp_aes_gcm.ib_flags = attr->flags;
3927		- memcpy(&accel_attrs.keymat.aes_gcm.aes_key, &aes_gcm->aes_key,
3928		- sizeof(accel_attrs.keymat.aes_gcm.aes_key));
3929		- accel_attrs.keymat.aes_gcm.key_len = aes_gcm->key_len * 8;
3930		- memcpy(&accel_attrs.keymat.aes_gcm.salt, &aes_gcm->salt,
3931		- sizeof(accel_attrs.keymat.aes_gcm.salt));
3932		- memcpy(&accel_attrs.keymat.aes_gcm.seq_iv, &aes_gcm->iv,
3933		- sizeof(accel_attrs.keymat.aes_gcm.seq_iv));
3934		- accel_attrs.keymat.aes_gcm.icv_len = aes_gcm->icv_len * 8;
3935		- accel_attrs.keymat.aes_gcm.iv_algo = MLX5_ACCEL_ESP_AES_GCM_IV_ALGO_SEQ;
3936		- accel_attrs.keymat_type = MLX5_ACCEL_ESP_KEYMAT_AES_GCM;
3937		-
3938		- accel_attrs.esn = attr->esn;
3939		- if (attr->flags & IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED)
3940		- accel_attrs.flags \|= MLX5_ACCEL_ESP_FLAGS_ESN_TRIGGERED;
3941		- if (attr->flags & IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ESN_NEW_WINDOW)
3942		- accel_attrs.flags \|= MLX5_ACCEL_ESP_FLAGS_ESN_STATE_OVERLAP;
3943		-
3944		- if (attr->flags & IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ENCRYPT)
3945		- accel_attrs.action \|= MLX5_ACCEL_ESP_ACTION_ENCRYPT;
3946		-
3947		- action->esp_aes_gcm.ctx =
3948		- mlx5_accel_esp_create_xfrm(mdev->mdev, &accel_attrs, flags);
3949		- if (IS_ERR(action->esp_aes_gcm.ctx)) {
3950		- err = PTR_ERR(action->esp_aes_gcm.ctx);
3951		- goto err_parse;
3952		- }
3953		-
3954		- action->esp_aes_gcm.ib_flags = attr->flags;
3955		-
3956		- return &action->ib_action;
3957		-
3958		-err_parse:
3959		- kfree(action);
3960		- return ERR_PTR(err);
3961		-}
3962		-
3963		-static int
3964		-mlx5_ib_modify_flow_action_esp(struct ib_flow_action *action,
3965		- const struct ib_flow_action_attrs_esp *attr,
3966		- struct uverbs_attr_bundle *attrs)
3967		-{
3968		- struct mlx5_ib_flow_action *maction = to_mflow_act(action);
3969		- struct mlx5_accel_esp_xfrm_attrs accel_attrs;
3970		- int err = 0;
3971		-
3972		- if (attr->keymat \|\| attr->replay \|\| attr->encap \|\|
3973		- attr->spi \|\| attr->seq \|\| attr->tfc_pad \|\|
3974		- attr->hard_limit_pkts \|\|
3975		- (attr->flags & ~(IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED \|
3976		- IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS \|
3977		- IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ESN_NEW_WINDOW)))
3978		- return -EOPNOTSUPP;
3979		-
3980		- /* Only the ESN value or the MLX5_ACCEL_ESP_FLAGS_ESN_STATE_OVERLAP can
3981		- * be modified.
3982		- */
3983		- if (!(maction->esp_aes_gcm.ib_flags &
3984		- IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED) &&
3985		- attr->flags & (IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED \|
3986		- IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ESN_NEW_WINDOW))
3987		- return -EINVAL;
3988		-
3989		- memcpy(&accel_attrs, &maction->esp_aes_gcm.ctx->attrs,
3990		- sizeof(accel_attrs));
3991		-
3992		- accel_attrs.esn = attr->esn;
3993		- if (attr->flags & IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ESN_NEW_WINDOW)
3994		- accel_attrs.flags \|= MLX5_ACCEL_ESP_FLAGS_ESN_STATE_OVERLAP;
3995		- else
3996		- accel_attrs.flags &= ~MLX5_ACCEL_ESP_FLAGS_ESN_STATE_OVERLAP;
3997		-
3998		- err = mlx5_accel_esp_modify_xfrm(maction->esp_aes_gcm.ctx,
3999		- &accel_attrs);
4000		- if (err)
4001		- return err;
4002		-
4003		- maction->esp_aes_gcm.ib_flags &=
4004		- ~IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ESN_NEW_WINDOW;
4005		- maction->esp_aes_gcm.ib_flags \|=
4006		- attr->flags & IB_UVERBS_FLOW_ACTION_ESP_FLAGS_ESN_NEW_WINDOW;
4007		-
4008		- return 0;
4009		-}
4010		-
4011		-static int mlx5_ib_destroy_flow_action(struct ib_flow_action *action)
4012		-{
4013		- struct mlx5_ib_flow_action *maction = to_mflow_act(action);
4014		-
4015		- switch (action->type) {
4016		- case IB_FLOW_ACTION_ESP:
4017		- /*
4018		- * We only support aes_gcm by now, so we implicitly know this is
4019		- * the underline crypto.
4020		- */
4021		- mlx5_accel_esp_destroy_xfrm(maction->esp_aes_gcm.ctx);
4022		- break;
4023		- default:
4024		- WARN_ON(true);
4025		- break;
4026		- }
4027		-
4028		- kfree(maction);
4029		- return 0;
	2586	+ return mlx5_cmd_dealloc_pd(mdev->mdev, mpd->pdn, mpd->uid);
4030	2587	}
4031	2588
4032	2589	static int mlx5_ib_mcg_attach(struct ib_qp ibqp, union ib_gid gid, u16 lid)
..	..	@@ -4034,13 +2591,17 @@
4034	2591	struct mlx5_ib_dev *dev = to_mdev(ibqp->device);
4035	2592	struct mlx5_ib_qp *mqp = to_mqp(ibqp);
4036	2593	int err;
	2594	+ u16 uid;
4037	2595
4038		- if (mqp->flags & MLX5_IB_QP_UNDERLAY) {
	2596	+ uid = ibqp->pd ?
	2597	+ to_mpd(ibqp->pd)->uid : 0;
	2598	+
	2599	+ if (mqp->flags & IB_QP_CREATE_SOURCE_QPN) {
4039	2600	mlx5_ib_dbg(dev, "Attaching a multi cast group to underlay QP is not supported\n");
4040	2601	return -EOPNOTSUPP;
4041	2602	}
4042	2603
4043		- err = mlx5_core_attach_mcg(dev->mdev, gid, ibqp->qp_num);
	2604	+ err = mlx5_cmd_attach_mcg(dev->mdev, gid, ibqp->qp_num, uid);
4044	2605	if (err)
4045	2606	mlx5_ib_warn(dev, "failed attaching QPN 0x%x, MGID %pI6\n",
4046	2607	ibqp->qp_num, gid->raw);
..	..	@@ -4052,8 +2613,11 @@
4052	2613	{
4053	2614	struct mlx5_ib_dev *dev = to_mdev(ibqp->device);
4054	2615	int err;
	2616	+ u16 uid;
4055	2617
4056		- err = mlx5_core_detach_mcg(dev->mdev, gid, ibqp->qp_num);
	2618	+ uid = ibqp->pd ?
	2619	+ to_mpd(ibqp->pd)->uid : 0;
	2620	+ err = mlx5_cmd_detach_mcg(dev->mdev, gid, ibqp->qp_num, uid);
4057	2621	if (err)
4058	2622	mlx5_ib_warn(dev, "failed detaching QPN 0x%x, MGID %pI6\n",
4059	2623	ibqp->qp_num, gid->raw);
..	..	@@ -4074,61 +2638,68 @@
4074	2638	return mlx5_query_node_guid(dev, &dev->ib_dev.node_guid);
4075	2639	}
4076	2640
4077		-static ssize_t show_fw_pages(struct device device, struct device_attribute attr,
4078		- char *buf)
	2641	+static ssize_t fw_pages_show(struct device *device,
	2642	+ struct device_attribute attr, char buf)
4079	2643	{
4080	2644	struct mlx5_ib_dev *dev =
4081		- container_of(device, struct mlx5_ib_dev, ib_dev.dev);
	2645	+ rdma_device_to_drv_device(device, struct mlx5_ib_dev, ib_dev);
4082	2646
4083	2647	return sprintf(buf, "%d\n", dev->mdev->priv.fw_pages);
4084	2648	}
	2649	+static DEVICE_ATTR_RO(fw_pages);
4085	2650
4086		-static ssize_t show_reg_pages(struct device *device,
	2651	+static ssize_t reg_pages_show(struct device *device,
4087	2652	struct device_attribute attr, char buf)
4088	2653	{
4089	2654	struct mlx5_ib_dev *dev =
4090		- container_of(device, struct mlx5_ib_dev, ib_dev.dev);
	2655	+ rdma_device_to_drv_device(device, struct mlx5_ib_dev, ib_dev);
4091	2656
4092	2657	return sprintf(buf, "%d\n", atomic_read(&dev->mdev->priv.reg_pages));
4093	2658	}
	2659	+static DEVICE_ATTR_RO(reg_pages);
4094	2660
4095		-static ssize_t show_hca(struct device device, struct device_attribute attr,
4096		- char *buf)
	2661	+static ssize_t hca_type_show(struct device *device,
	2662	+ struct device_attribute attr, char buf)
4097	2663	{
4098	2664	struct mlx5_ib_dev *dev =
4099		- container_of(device, struct mlx5_ib_dev, ib_dev.dev);
	2665	+ rdma_device_to_drv_device(device, struct mlx5_ib_dev, ib_dev);
	2666	+
4100	2667	return sprintf(buf, "MT%d\n", dev->mdev->pdev->device);
4101	2668	}
	2669	+static DEVICE_ATTR_RO(hca_type);
4102	2670
4103		-static ssize_t show_rev(struct device device, struct device_attribute attr,
4104		- char *buf)
	2671	+static ssize_t hw_rev_show(struct device *device,
	2672	+ struct device_attribute attr, char buf)
4105	2673	{
4106	2674	struct mlx5_ib_dev *dev =
4107		- container_of(device, struct mlx5_ib_dev, ib_dev.dev);
	2675	+ rdma_device_to_drv_device(device, struct mlx5_ib_dev, ib_dev);
	2676	+
4108	2677	return sprintf(buf, "%x\n", dev->mdev->rev_id);
4109	2678	}
	2679	+static DEVICE_ATTR_RO(hw_rev);
4110	2680
4111		-static ssize_t show_board(struct device device, struct device_attribute attr,
4112		- char *buf)
	2681	+static ssize_t board_id_show(struct device *device,
	2682	+ struct device_attribute attr, char buf)
4113	2683	{
4114	2684	struct mlx5_ib_dev *dev =
4115		- container_of(device, struct mlx5_ib_dev, ib_dev.dev);
	2685	+ rdma_device_to_drv_device(device, struct mlx5_ib_dev, ib_dev);
	2686	+
4116	2687	return sprintf(buf, "%.*s\n", MLX5_BOARD_ID_LEN,
4117	2688	dev->mdev->board_id);
4118	2689	}
	2690	+static DEVICE_ATTR_RO(board_id);
4119	2691
4120		-static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
4121		-static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
4122		-static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
4123		-static DEVICE_ATTR(fw_pages, S_IRUGO, show_fw_pages, NULL);
4124		-static DEVICE_ATTR(reg_pages, S_IRUGO, show_reg_pages, NULL);
	2692	+static struct attribute *mlx5_class_attributes[] = {
	2693	+ &dev_attr_hw_rev.attr,
	2694	+ &dev_attr_hca_type.attr,
	2695	+ &dev_attr_board_id.attr,
	2696	+ &dev_attr_fw_pages.attr,
	2697	+ &dev_attr_reg_pages.attr,
	2698	+ NULL,
	2699	+};
4125	2700
4126		-static struct device_attribute *mlx5_class_attributes[] = {
4127		- &dev_attr_hw_rev,
4128		- &dev_attr_hca_type,
4129		- &dev_attr_board_id,
4130		- &dev_attr_fw_pages,
4131		- &dev_attr_reg_pages,
	2701	+static const struct attribute_group mlx5_attr_group = {
	2702	+ .attrs = mlx5_class_attributes,
4132	2703	};
4133	2704
4134	2705	static void pkey_change_handler(struct work_struct *work)
..	..	@@ -4137,9 +2708,7 @@
4137	2708	container_of(work, struct mlx5_ib_port_resources,
4138	2709	pkey_change_work);
4139	2710
4140		- mutex_lock(&ports->devr->mutex);
4141	2711	mlx5_ib_gsi_pkey_change(ports->gsi);
4142		- mutex_unlock(&ports->devr->mutex);
4143	2712	}
4144	2713
4145	2714	static void mlx5_ib_handle_internal_error(struct mlx5_ib_dev *ibdev)
..	..	@@ -4196,7 +2765,7 @@
4196	2765	* lock/unlock above locks Now need to arm all involved CQs.
4197	2766	*/
4198	2767	list_for_each_entry(mcq, &cq_armed_list, reset_notify) {
4199		- mcq->comp(mcq);
	2768	+ mcq->comp(mcq, NULL);
4200	2769	}
4201	2770	spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
4202	2771	}
..	..	@@ -4211,14 +2780,74 @@
4211	2780	atomic_inc(&delay_drop->events_cnt);
4212	2781
4213	2782	mutex_lock(&delay_drop->lock);
4214		- err = mlx5_core_set_delay_drop(delay_drop->dev->mdev,
4215		- delay_drop->timeout);
	2783	+ err = mlx5_core_set_delay_drop(delay_drop->dev, delay_drop->timeout);
4216	2784	if (err) {
4217	2785	mlx5_ib_warn(delay_drop->dev, "Failed to set delay drop, timeout=%u\n",
4218	2786	delay_drop->timeout);
4219	2787	delay_drop->activate = false;
4220	2788	}
4221	2789	mutex_unlock(&delay_drop->lock);
	2790	+}
	2791	+
	2792	+static void handle_general_event(struct mlx5_ib_dev ibdev, struct mlx5_eqe eqe,
	2793	+ struct ib_event *ibev)
	2794	+{
	2795	+ u8 port = (eqe->data.port.port >> 4) & 0xf;
	2796	+
	2797	+ switch (eqe->sub_type) {
	2798	+ case MLX5_GENERAL_SUBTYPE_DELAY_DROP_TIMEOUT:
	2799	+ if (mlx5_ib_port_link_layer(&ibdev->ib_dev, port) ==
	2800	+ IB_LINK_LAYER_ETHERNET)
	2801	+ schedule_work(&ibdev->delay_drop.delay_drop_work);
	2802	+ break;
	2803	+ default: /* do nothing */
	2804	+ return;
	2805	+ }
	2806	+}
	2807	+
	2808	+static int handle_port_change(struct mlx5_ib_dev ibdev, struct mlx5_eqe eqe,
	2809	+ struct ib_event *ibev)
	2810	+{
	2811	+ u8 port = (eqe->data.port.port >> 4) & 0xf;
	2812	+
	2813	+ ibev->element.port_num = port;
	2814	+
	2815	+ switch (eqe->sub_type) {
	2816	+ case MLX5_PORT_CHANGE_SUBTYPE_ACTIVE:
	2817	+ case MLX5_PORT_CHANGE_SUBTYPE_DOWN:
	2818	+ case MLX5_PORT_CHANGE_SUBTYPE_INITIALIZED:
	2819	+ /* In RoCE, port up/down events are handled in
	2820	+ * mlx5_netdev_event().
	2821	+ */
	2822	+ if (mlx5_ib_port_link_layer(&ibdev->ib_dev, port) ==
	2823	+ IB_LINK_LAYER_ETHERNET)
	2824	+ return -EINVAL;
	2825	+
	2826	+ ibev->event = (eqe->sub_type == MLX5_PORT_CHANGE_SUBTYPE_ACTIVE) ?
	2827	+ IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
	2828	+ break;
	2829	+
	2830	+ case MLX5_PORT_CHANGE_SUBTYPE_LID:
	2831	+ ibev->event = IB_EVENT_LID_CHANGE;
	2832	+ break;
	2833	+
	2834	+ case MLX5_PORT_CHANGE_SUBTYPE_PKEY:
	2835	+ ibev->event = IB_EVENT_PKEY_CHANGE;
	2836	+ schedule_work(&ibdev->devr.ports[port - 1].pkey_change_work);
	2837	+ break;
	2838	+
	2839	+ case MLX5_PORT_CHANGE_SUBTYPE_GUID:
	2840	+ ibev->event = IB_EVENT_GID_CHANGE;
	2841	+ break;
	2842	+
	2843	+ case MLX5_PORT_CHANGE_SUBTYPE_CLIENT_REREG:
	2844	+ ibev->event = IB_EVENT_CLIENT_REREGISTER;
	2845	+ break;
	2846	+ default:
	2847	+ return -EINVAL;
	2848	+ }
	2849	+
	2850	+ return 0;
4222	2851	}
4223	2852
4224	2853	static void mlx5_ib_handle_event(struct work_struct *_work)
..	..	@@ -4228,65 +2857,37 @@
4228	2857	struct mlx5_ib_dev *ibdev;
4229	2858	struct ib_event ibev;
4230	2859	bool fatal = false;
4231		- u8 port = (u8)work->param;
4232	2860
4233		- if (mlx5_core_is_mp_slave(work->dev)) {
4234		- ibdev = mlx5_ib_get_ibdev_from_mpi(work->context);
	2861	+ if (work->is_slave) {
	2862	+ ibdev = mlx5_ib_get_ibdev_from_mpi(work->mpi);
4235	2863	if (!ibdev)
4236	2864	goto out;
4237	2865	} else {
4238		- ibdev = work->context;
	2866	+ ibdev = work->dev;
4239	2867	}
4240	2868
4241	2869	switch (work->event) {
4242	2870	case MLX5_DEV_EVENT_SYS_ERROR:
4243	2871	ibev.event = IB_EVENT_DEVICE_FATAL;
4244	2872	mlx5_ib_handle_internal_error(ibdev);
	2873	+ ibev.element.port_num = (u8)(unsigned long)work->param;
4245	2874	fatal = true;
4246	2875	break;
4247		-
4248		- case MLX5_DEV_EVENT_PORT_UP:
4249		- case MLX5_DEV_EVENT_PORT_DOWN:
4250		- case MLX5_DEV_EVENT_PORT_INITIALIZED:
4251		- /* In RoCE, port up/down events are handled in
4252		- * mlx5_netdev_event().
4253		- */
4254		- if (mlx5_ib_port_link_layer(&ibdev->ib_dev, port) ==
4255		- IB_LINK_LAYER_ETHERNET)
	2876	+ case MLX5_EVENT_TYPE_PORT_CHANGE:
	2877	+ if (handle_port_change(ibdev, work->param, &ibev))
4256	2878	goto out;
4257		-
4258		- ibev.event = (work->event == MLX5_DEV_EVENT_PORT_UP) ?
4259		- IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
4260	2879	break;
4261		-
4262		- case MLX5_DEV_EVENT_LID_CHANGE:
4263		- ibev.event = IB_EVENT_LID_CHANGE;
4264		- break;
4265		-
4266		- case MLX5_DEV_EVENT_PKEY_CHANGE:
4267		- ibev.event = IB_EVENT_PKEY_CHANGE;
4268		- schedule_work(&ibdev->devr.ports[port - 1].pkey_change_work);
4269		- break;
4270		-
4271		- case MLX5_DEV_EVENT_GUID_CHANGE:
4272		- ibev.event = IB_EVENT_GID_CHANGE;
4273		- break;
4274		-
4275		- case MLX5_DEV_EVENT_CLIENT_REREG:
4276		- ibev.event = IB_EVENT_CLIENT_REREGISTER;
4277		- break;
4278		- case MLX5_DEV_EVENT_DELAY_DROP_TIMEOUT:
4279		- schedule_work(&ibdev->delay_drop.delay_drop_work);
4280		- goto out;
	2880	+ case MLX5_EVENT_TYPE_GENERAL_EVENT:
	2881	+ handle_general_event(ibdev, work->param, &ibev);
	2882	+ fallthrough;
4281	2883	default:
4282	2884	goto out;
4283	2885	}
4284	2886
4285		- ibev.device = &ibdev->ib_dev;
4286		- ibev.element.port_num = port;
	2887	+ ibev.device = &ibdev->ib_dev;
4287	2888
4288		- if (!rdma_is_port_valid(&ibdev->ib_dev, port)) {
4289		- mlx5_ib_warn(ibdev, "warning: event on port %d\n", port);
	2889	+ if (!rdma_is_port_valid(&ibdev->ib_dev, ibev.element.port_num)) {
	2890	+ mlx5_ib_warn(ibdev, "warning: event on port %d\n", ibev.element.port_num);
4290	2891	goto out;
4291	2892	}
4292	2893
..	..	@@ -4299,22 +2900,43 @@
4299	2900	kfree(work);
4300	2901	}
4301	2902
4302		-static void mlx5_ib_event(struct mlx5_core_dev dev, void context,
4303		- enum mlx5_dev_event event, unsigned long param)
	2903	+static int mlx5_ib_event(struct notifier_block *nb,
	2904	+ unsigned long event, void *param)
4304	2905	{
4305	2906	struct mlx5_ib_event_work *work;
4306	2907
4307	2908	work = kmalloc(sizeof(*work), GFP_ATOMIC);
4308	2909	if (!work)
4309		- return;
	2910	+ return NOTIFY_DONE;
4310	2911
4311	2912	INIT_WORK(&work->work, mlx5_ib_handle_event);
4312		- work->dev = dev;
	2913	+ work->dev = container_of(nb, struct mlx5_ib_dev, mdev_events);
	2914	+ work->is_slave = false;
4313	2915	work->param = param;
4314		- work->context = context;
4315	2916	work->event = event;
4316	2917
4317	2918	queue_work(mlx5_ib_event_wq, &work->work);
	2919	+
	2920	+ return NOTIFY_OK;
	2921	+}
	2922	+
	2923	+static int mlx5_ib_event_slave_port(struct notifier_block *nb,
	2924	+ unsigned long event, void *param)
	2925	+{
	2926	+ struct mlx5_ib_event_work *work;
	2927	+
	2928	+ work = kmalloc(sizeof(*work), GFP_ATOMIC);
	2929	+ if (!work)
	2930	+ return NOTIFY_DONE;
	2931	+
	2932	+ INIT_WORK(&work->work, mlx5_ib_handle_event);
	2933	+ work->mpi = container_of(nb, struct mlx5_ib_multiport_info, mdev_events);
	2934	+ work->is_slave = true;
	2935	+ work->param = param;
	2936	+ work->event = event;
	2937	+ queue_work(mlx5_ib_event_wq, &work->work);
	2938	+
	2939	+ return NOTIFY_OK;
4318	2940	}
4319	2941
4320	2942	static int set_has_smi_cap(struct mlx5_ib_dev *dev)
..	..	@@ -4323,7 +2945,7 @@
4323	2945	int err;
4324	2946	int port;
4325	2947
4326		- for (port = 1; port <= dev->num_ports; port++) {
	2948	+ for (port = 1; port <= ARRAY_SIZE(dev->mdev->port_caps); port++) {
4327	2949	dev->mdev->port_caps[port - 1].has_smi = false;
4328	2950	if (MLX5_CAP_GEN(dev->mdev, port_type) ==
4329	2951	MLX5_CAP_PORT_TYPE_IB) {
..	..	@@ -4354,14 +2976,13 @@
4354	2976	mlx5_query_ext_port_caps(dev, port);
4355	2977	}
4356	2978
4357		-static int get_port_caps(struct mlx5_ib_dev *dev, u8 port)
	2979	+static int __get_port_caps(struct mlx5_ib_dev *dev, u8 port)
4358	2980	{
4359	2981	struct ib_device_attr *dprops = NULL;
4360	2982	struct ib_port_attr *pprops = NULL;
4361	2983	int err = -ENOMEM;
4362		- struct ib_udata uhw = {.inlen = 0, .outlen = 0};
4363	2984
4364		- pprops = kmalloc(sizeof(*pprops), GFP_KERNEL);
	2985	+ pprops = kzalloc(sizeof(*pprops), GFP_KERNEL);
4365	2986	if (!pprops)
4366	2987	goto out;
4367	2988
..	..	@@ -4369,17 +2990,12 @@
4369	2990	if (!dprops)
4370	2991	goto out;
4371	2992
4372		- err = set_has_smi_cap(dev);
4373		- if (err)
4374		- goto out;
4375		-
4376		- err = mlx5_ib_query_device(&dev->ib_dev, dprops, &uhw);
	2993	+ err = mlx5_ib_query_device(&dev->ib_dev, dprops, NULL);
4377	2994	if (err) {
4378	2995	mlx5_ib_warn(dev, "query_device failed %d\n", err);
4379	2996	goto out;
4380	2997	}
4381	2998
4382		- memset(pprops, 0, sizeof(*pprops));
4383	2999	err = mlx5_ib_query_port(&dev->ib_dev, port, pprops);
4384	3000	if (err) {
4385	3001	mlx5_ib_warn(dev, "query_port %d failed %d\n",
..	..	@@ -4401,7 +3017,1377 @@
4401	3017	return err;
4402	3018	}
4403	3019
4404		-static void destroy_umrc_res(struct mlx5_ib_dev *dev)
	3020	+static int get_port_caps(struct mlx5_ib_dev *dev, u8 port)
	3021	+{
	3022	+ /* For representors use port 1, is this is the only native
	3023	+ * port
	3024	+ */
	3025	+ if (dev->is_rep)
	3026	+ return __get_port_caps(dev, 1);
	3027	+ return __get_port_caps(dev, port);
	3028	+}
	3029	+
	3030	+static u8 mlx5_get_umr_fence(u8 umr_fence_cap)
	3031	+{
	3032	+ switch (umr_fence_cap) {
	3033	+ case MLX5_CAP_UMR_FENCE_NONE:
	3034	+ return MLX5_FENCE_MODE_NONE;
	3035	+ case MLX5_CAP_UMR_FENCE_SMALL:
	3036	+ return MLX5_FENCE_MODE_INITIATOR_SMALL;
	3037	+ default:
	3038	+ return MLX5_FENCE_MODE_STRONG_ORDERING;
	3039	+ }
	3040	+}
	3041	+
	3042	+static int mlx5_ib_dev_res_init(struct mlx5_ib_dev *dev)
	3043	+{
	3044	+ struct mlx5_ib_resources *devr = &dev->devr;
	3045	+ struct ib_srq_init_attr attr;
	3046	+ struct ib_device *ibdev;
	3047	+ struct ib_cq_init_attr cq_attr = {.cqe = 1};
	3048	+ int port;
	3049	+ int ret = 0;
	3050	+
	3051	+ ibdev = &dev->ib_dev;
	3052	+
	3053	+ if (!MLX5_CAP_GEN(dev->mdev, xrc))
	3054	+ return -EOPNOTSUPP;
	3055	+
	3056	+ mutex_init(&devr->mutex);
	3057	+
	3058	+ devr->p0 = rdma_zalloc_drv_obj(ibdev, ib_pd);
	3059	+ if (!devr->p0)
	3060	+ return -ENOMEM;
	3061	+
	3062	+ devr->p0->device = ibdev;
	3063	+ devr->p0->uobject = NULL;
	3064	+ atomic_set(&devr->p0->usecnt, 0);
	3065	+
	3066	+ ret = mlx5_ib_alloc_pd(devr->p0, NULL);
	3067	+ if (ret)
	3068	+ goto error0;
	3069	+
	3070	+ devr->c0 = rdma_zalloc_drv_obj(ibdev, ib_cq);
	3071	+ if (!devr->c0) {
	3072	+ ret = -ENOMEM;
	3073	+ goto error1;
	3074	+ }
	3075	+
	3076	+ devr->c0->device = &dev->ib_dev;
	3077	+ atomic_set(&devr->c0->usecnt, 0);
	3078	+
	3079	+ ret = mlx5_ib_create_cq(devr->c0, &cq_attr, NULL);
	3080	+ if (ret)
	3081	+ goto err_create_cq;
	3082	+
	3083	+ ret = mlx5_cmd_xrcd_alloc(dev->mdev, &devr->xrcdn0, 0);
	3084	+ if (ret)
	3085	+ goto error2;
	3086	+
	3087	+ ret = mlx5_cmd_xrcd_alloc(dev->mdev, &devr->xrcdn1, 0);
	3088	+ if (ret)
	3089	+ goto error3;
	3090	+
	3091	+ memset(&attr, 0, sizeof(attr));
	3092	+ attr.attr.max_sge = 1;
	3093	+ attr.attr.max_wr = 1;
	3094	+ attr.srq_type = IB_SRQT_XRC;
	3095	+ attr.ext.cq = devr->c0;
	3096	+
	3097	+ devr->s0 = rdma_zalloc_drv_obj(ibdev, ib_srq);
	3098	+ if (!devr->s0) {
	3099	+ ret = -ENOMEM;
	3100	+ goto error4;
	3101	+ }
	3102	+
	3103	+ devr->s0->device = &dev->ib_dev;
	3104	+ devr->s0->pd = devr->p0;
	3105	+ devr->s0->srq_type = IB_SRQT_XRC;
	3106	+ devr->s0->ext.cq = devr->c0;
	3107	+ ret = mlx5_ib_create_srq(devr->s0, &attr, NULL);
	3108	+ if (ret)
	3109	+ goto err_create;
	3110	+
	3111	+ atomic_inc(&devr->s0->ext.cq->usecnt);
	3112	+ atomic_inc(&devr->p0->usecnt);
	3113	+ atomic_set(&devr->s0->usecnt, 0);
	3114	+
	3115	+ memset(&attr, 0, sizeof(attr));
	3116	+ attr.attr.max_sge = 1;
	3117	+ attr.attr.max_wr = 1;
	3118	+ attr.srq_type = IB_SRQT_BASIC;
	3119	+ devr->s1 = rdma_zalloc_drv_obj(ibdev, ib_srq);
	3120	+ if (!devr->s1) {
	3121	+ ret = -ENOMEM;
	3122	+ goto error5;
	3123	+ }
	3124	+
	3125	+ devr->s1->device = &dev->ib_dev;
	3126	+ devr->s1->pd = devr->p0;
	3127	+ devr->s1->srq_type = IB_SRQT_BASIC;
	3128	+ devr->s1->ext.cq = devr->c0;
	3129	+
	3130	+ ret = mlx5_ib_create_srq(devr->s1, &attr, NULL);
	3131	+ if (ret)
	3132	+ goto error6;
	3133	+
	3134	+ atomic_inc(&devr->p0->usecnt);
	3135	+ atomic_set(&devr->s1->usecnt, 0);
	3136	+
	3137	+ for (port = 0; port < ARRAY_SIZE(devr->ports); ++port)
	3138	+ INIT_WORK(&devr->ports[port].pkey_change_work,
	3139	+ pkey_change_handler);
	3140	+
	3141	+ return 0;
	3142	+
	3143	+error6:
	3144	+ kfree(devr->s1);
	3145	+error5:
	3146	+ mlx5_ib_destroy_srq(devr->s0, NULL);
	3147	+err_create:
	3148	+ kfree(devr->s0);
	3149	+error4:
	3150	+ mlx5_cmd_xrcd_dealloc(dev->mdev, devr->xrcdn1, 0);
	3151	+error3:
	3152	+ mlx5_cmd_xrcd_dealloc(dev->mdev, devr->xrcdn0, 0);
	3153	+error2:
	3154	+ mlx5_ib_destroy_cq(devr->c0, NULL);
	3155	+err_create_cq:
	3156	+ kfree(devr->c0);
	3157	+error1:
	3158	+ mlx5_ib_dealloc_pd(devr->p0, NULL);
	3159	+error0:
	3160	+ kfree(devr->p0);
	3161	+ return ret;
	3162	+}
	3163	+
	3164	+static void mlx5_ib_dev_res_cleanup(struct mlx5_ib_dev *dev)
	3165	+{
	3166	+ struct mlx5_ib_resources *devr = &dev->devr;
	3167	+ int port;
	3168	+
	3169	+ mlx5_ib_destroy_srq(devr->s1, NULL);
	3170	+ kfree(devr->s1);
	3171	+ mlx5_ib_destroy_srq(devr->s0, NULL);
	3172	+ kfree(devr->s0);
	3173	+ mlx5_cmd_xrcd_dealloc(dev->mdev, devr->xrcdn1, 0);
	3174	+ mlx5_cmd_xrcd_dealloc(dev->mdev, devr->xrcdn0, 0);
	3175	+ mlx5_ib_destroy_cq(devr->c0, NULL);
	3176	+ kfree(devr->c0);
	3177	+ mlx5_ib_dealloc_pd(devr->p0, NULL);
	3178	+ kfree(devr->p0);
	3179	+
	3180	+ /* Make sure no change P_Key work items are still executing */
	3181	+ for (port = 0; port < ARRAY_SIZE(devr->ports); ++port)
	3182	+ cancel_work_sync(&devr->ports[port].pkey_change_work);
	3183	+}
	3184	+
	3185	+static u32 get_core_cap_flags(struct ib_device *ibdev,
	3186	+ struct mlx5_hca_vport_context *rep)
	3187	+{
	3188	+ struct mlx5_ib_dev *dev = to_mdev(ibdev);
	3189	+ enum rdma_link_layer ll = mlx5_ib_port_link_layer(ibdev, 1);
	3190	+ u8 l3_type_cap = MLX5_CAP_ROCE(dev->mdev, l3_type);
	3191	+ u8 roce_version_cap = MLX5_CAP_ROCE(dev->mdev, roce_version);
	3192	+ bool raw_support = !mlx5_core_mp_enabled(dev->mdev);
	3193	+ u32 ret = 0;
	3194	+
	3195	+ if (rep->grh_required)
	3196	+ ret \|= RDMA_CORE_CAP_IB_GRH_REQUIRED;
	3197	+
	3198	+ if (ll == IB_LINK_LAYER_INFINIBAND)
	3199	+ return ret \| RDMA_CORE_PORT_IBA_IB;
	3200	+
	3201	+ if (raw_support)
	3202	+ ret \|= RDMA_CORE_PORT_RAW_PACKET;
	3203	+
	3204	+ if (!(l3_type_cap & MLX5_ROCE_L3_TYPE_IPV4_CAP))
	3205	+ return ret;
	3206	+
	3207	+ if (!(l3_type_cap & MLX5_ROCE_L3_TYPE_IPV6_CAP))
	3208	+ return ret;
	3209	+
	3210	+ if (roce_version_cap & MLX5_ROCE_VERSION_1_CAP)
	3211	+ ret \|= RDMA_CORE_PORT_IBA_ROCE;
	3212	+
	3213	+ if (roce_version_cap & MLX5_ROCE_VERSION_2_CAP)
	3214	+ ret \|= RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
	3215	+
	3216	+ return ret;
	3217	+}
	3218	+
	3219	+static int mlx5_port_immutable(struct ib_device *ibdev, u8 port_num,
	3220	+ struct ib_port_immutable *immutable)
	3221	+{
	3222	+ struct ib_port_attr attr;
	3223	+ struct mlx5_ib_dev *dev = to_mdev(ibdev);
	3224	+ enum rdma_link_layer ll = mlx5_ib_port_link_layer(ibdev, port_num);
	3225	+ struct mlx5_hca_vport_context rep = {0};
	3226	+ int err;
	3227	+
	3228	+ err = ib_query_port(ibdev, port_num, &attr);
	3229	+ if (err)
	3230	+ return err;
	3231	+
	3232	+ if (ll == IB_LINK_LAYER_INFINIBAND) {
	3233	+ err = mlx5_query_hca_vport_context(dev->mdev, 0, port_num, 0,
	3234	+ &rep);
	3235	+ if (err)
	3236	+ return err;
	3237	+ }
	3238	+
	3239	+ immutable->pkey_tbl_len = attr.pkey_tbl_len;
	3240	+ immutable->gid_tbl_len = attr.gid_tbl_len;
	3241	+ immutable->core_cap_flags = get_core_cap_flags(ibdev, &rep);
	3242	+ immutable->max_mad_size = IB_MGMT_MAD_SIZE;
	3243	+
	3244	+ return 0;
	3245	+}
	3246	+
	3247	+static int mlx5_port_rep_immutable(struct ib_device *ibdev, u8 port_num,
	3248	+ struct ib_port_immutable *immutable)
	3249	+{
	3250	+ struct ib_port_attr attr;
	3251	+ int err;
	3252	+
	3253	+ immutable->core_cap_flags = RDMA_CORE_PORT_RAW_PACKET;
	3254	+
	3255	+ err = ib_query_port(ibdev, port_num, &attr);
	3256	+ if (err)
	3257	+ return err;
	3258	+
	3259	+ immutable->pkey_tbl_len = attr.pkey_tbl_len;
	3260	+ immutable->gid_tbl_len = attr.gid_tbl_len;
	3261	+ immutable->core_cap_flags = RDMA_CORE_PORT_RAW_PACKET;
	3262	+
	3263	+ return 0;
	3264	+}
	3265	+
	3266	+static void get_dev_fw_str(struct ib_device ibdev, char str)
	3267	+{
	3268	+ struct mlx5_ib_dev *dev =
	3269	+ container_of(ibdev, struct mlx5_ib_dev, ib_dev);
	3270	+ snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%04d",
	3271	+ fw_rev_maj(dev->mdev), fw_rev_min(dev->mdev),
	3272	+ fw_rev_sub(dev->mdev));
	3273	+}
	3274	+
	3275	+static int mlx5_eth_lag_init(struct mlx5_ib_dev *dev)
	3276	+{
	3277	+ struct mlx5_core_dev *mdev = dev->mdev;
	3278	+ struct mlx5_flow_namespace *ns = mlx5_get_flow_namespace(mdev,
	3279	+ MLX5_FLOW_NAMESPACE_LAG);
	3280	+ struct mlx5_flow_table *ft;
	3281	+ int err;
	3282	+
	3283	+ if (!ns \|\| !mlx5_lag_is_roce(mdev))
	3284	+ return 0;
	3285	+
	3286	+ err = mlx5_cmd_create_vport_lag(mdev);
	3287	+ if (err)
	3288	+ return err;
	3289	+
	3290	+ ft = mlx5_create_lag_demux_flow_table(ns, 0, 0);
	3291	+ if (IS_ERR(ft)) {
	3292	+ err = PTR_ERR(ft);
	3293	+ goto err_destroy_vport_lag;
	3294	+ }
	3295	+
	3296	+ dev->flow_db->lag_demux_ft = ft;
	3297	+ dev->lag_active = true;
	3298	+ return 0;
	3299	+
	3300	+err_destroy_vport_lag:
	3301	+ mlx5_cmd_destroy_vport_lag(mdev);
	3302	+ return err;
	3303	+}
	3304	+
	3305	+static void mlx5_eth_lag_cleanup(struct mlx5_ib_dev *dev)
	3306	+{
	3307	+ struct mlx5_core_dev *mdev = dev->mdev;
	3308	+
	3309	+ if (dev->lag_active) {
	3310	+ dev->lag_active = false;
	3311	+
	3312	+ mlx5_destroy_flow_table(dev->flow_db->lag_demux_ft);
	3313	+ dev->flow_db->lag_demux_ft = NULL;
	3314	+
	3315	+ mlx5_cmd_destroy_vport_lag(mdev);
	3316	+ }
	3317	+}
	3318	+
	3319	+static int mlx5_add_netdev_notifier(struct mlx5_ib_dev *dev, u8 port_num)
	3320	+{
	3321	+ int err;
	3322	+
	3323	+ dev->port[port_num].roce.nb.notifier_call = mlx5_netdev_event;
	3324	+ err = register_netdevice_notifier(&dev->port[port_num].roce.nb);
	3325	+ if (err) {
	3326	+ dev->port[port_num].roce.nb.notifier_call = NULL;
	3327	+ return err;
	3328	+ }
	3329	+
	3330	+ return 0;
	3331	+}
	3332	+
	3333	+static void mlx5_remove_netdev_notifier(struct mlx5_ib_dev *dev, u8 port_num)
	3334	+{
	3335	+ if (dev->port[port_num].roce.nb.notifier_call) {
	3336	+ unregister_netdevice_notifier(&dev->port[port_num].roce.nb);
	3337	+ dev->port[port_num].roce.nb.notifier_call = NULL;
	3338	+ }
	3339	+}
	3340	+
	3341	+static int mlx5_enable_eth(struct mlx5_ib_dev *dev)
	3342	+{
	3343	+ int err;
	3344	+
	3345	+ err = mlx5_nic_vport_enable_roce(dev->mdev);
	3346	+ if (err)
	3347	+ return err;
	3348	+
	3349	+ err = mlx5_eth_lag_init(dev);
	3350	+ if (err)
	3351	+ goto err_disable_roce;
	3352	+
	3353	+ return 0;
	3354	+
	3355	+err_disable_roce:
	3356	+ mlx5_nic_vport_disable_roce(dev->mdev);
	3357	+
	3358	+ return err;
	3359	+}
	3360	+
	3361	+static void mlx5_disable_eth(struct mlx5_ib_dev *dev)
	3362	+{
	3363	+ mlx5_eth_lag_cleanup(dev);
	3364	+ mlx5_nic_vport_disable_roce(dev->mdev);
	3365	+}
	3366	+
	3367	+static int mlx5_ib_rn_get_params(struct ib_device *device, u8 port_num,
	3368	+ enum rdma_netdev_t type,
	3369	+ struct rdma_netdev_alloc_params *params)
	3370	+{
	3371	+ if (type != RDMA_NETDEV_IPOIB)
	3372	+ return -EOPNOTSUPP;
	3373	+
	3374	+ return mlx5_rdma_rn_get_params(to_mdev(device)->mdev, device, params);
	3375	+}
	3376	+
	3377	+static ssize_t delay_drop_timeout_read(struct file filp, char __user buf,
	3378	+ size_t count, loff_t *pos)
	3379	+{
	3380	+ struct mlx5_ib_delay_drop *delay_drop = filp->private_data;
	3381	+ char lbuf[20];
	3382	+ int len;
	3383	+
	3384	+ len = snprintf(lbuf, sizeof(lbuf), "%u\n", delay_drop->timeout);
	3385	+ return simple_read_from_buffer(buf, count, pos, lbuf, len);
	3386	+}
	3387	+
	3388	+static ssize_t delay_drop_timeout_write(struct file filp, const char __user buf,
	3389	+ size_t count, loff_t *pos)
	3390	+{
	3391	+ struct mlx5_ib_delay_drop *delay_drop = filp->private_data;
	3392	+ u32 timeout;
	3393	+ u32 var;
	3394	+
	3395	+ if (kstrtouint_from_user(buf, count, 0, &var))
	3396	+ return -EFAULT;
	3397	+
	3398	+ timeout = min_t(u32, roundup(var, 100), MLX5_MAX_DELAY_DROP_TIMEOUT_MS *
	3399	+ 1000);
	3400	+ if (timeout != var)
	3401	+ mlx5_ib_dbg(delay_drop->dev, "Round delay drop timeout to %u usec\n",
	3402	+ timeout);
	3403	+
	3404	+ delay_drop->timeout = timeout;
	3405	+
	3406	+ return count;
	3407	+}
	3408	+
	3409	+static const struct file_operations fops_delay_drop_timeout = {
	3410	+ .owner = THIS_MODULE,
	3411	+ .open = simple_open,
	3412	+ .write = delay_drop_timeout_write,
	3413	+ .read = delay_drop_timeout_read,
	3414	+};
	3415	+
	3416	+static void mlx5_ib_unbind_slave_port(struct mlx5_ib_dev *ibdev,
	3417	+ struct mlx5_ib_multiport_info *mpi)
	3418	+{
	3419	+ u8 port_num = mlx5_core_native_port_num(mpi->mdev) - 1;
	3420	+ struct mlx5_ib_port *port = &ibdev->port[port_num];
	3421	+ int comps;
	3422	+ int err;
	3423	+ int i;
	3424	+
	3425	+ lockdep_assert_held(&mlx5_ib_multiport_mutex);
	3426	+
	3427	+ mlx5_ib_cleanup_cong_debugfs(ibdev, port_num);
	3428	+
	3429	+ spin_lock(&port->mp.mpi_lock);
	3430	+ if (!mpi->ibdev) {
	3431	+ spin_unlock(&port->mp.mpi_lock);
	3432	+ return;
	3433	+ }
	3434	+
	3435	+ mpi->ibdev = NULL;
	3436	+
	3437	+ spin_unlock(&port->mp.mpi_lock);
	3438	+ if (mpi->mdev_events.notifier_call)
	3439	+ mlx5_notifier_unregister(mpi->mdev, &mpi->mdev_events);
	3440	+ mpi->mdev_events.notifier_call = NULL;
	3441	+ mlx5_remove_netdev_notifier(ibdev, port_num);
	3442	+ spin_lock(&port->mp.mpi_lock);
	3443	+
	3444	+ comps = mpi->mdev_refcnt;
	3445	+ if (comps) {
	3446	+ mpi->unaffiliate = true;
	3447	+ init_completion(&mpi->unref_comp);
	3448	+ spin_unlock(&port->mp.mpi_lock);
	3449	+
	3450	+ for (i = 0; i < comps; i++)
	3451	+ wait_for_completion(&mpi->unref_comp);
	3452	+
	3453	+ spin_lock(&port->mp.mpi_lock);
	3454	+ mpi->unaffiliate = false;
	3455	+ }
	3456	+
	3457	+ port->mp.mpi = NULL;
	3458	+
	3459	+ spin_unlock(&port->mp.mpi_lock);
	3460	+
	3461	+ err = mlx5_nic_vport_unaffiliate_multiport(mpi->mdev);
	3462	+
	3463	+ mlx5_ib_dbg(ibdev, "unaffiliated port %d\n", port_num + 1);
	3464	+ /* Log an error, still needed to cleanup the pointers and add
	3465	+ * it back to the list.
	3466	+ */
	3467	+ if (err)
	3468	+ mlx5_ib_err(ibdev, "Failed to unaffiliate port %u\n",
	3469	+ port_num + 1);
	3470	+
	3471	+ ibdev->port[port_num].roce.last_port_state = IB_PORT_DOWN;
	3472	+}
	3473	+
	3474	+static bool mlx5_ib_bind_slave_port(struct mlx5_ib_dev *ibdev,
	3475	+ struct mlx5_ib_multiport_info *mpi)
	3476	+{
	3477	+ u8 port_num = mlx5_core_native_port_num(mpi->mdev) - 1;
	3478	+ int err;
	3479	+
	3480	+ lockdep_assert_held(&mlx5_ib_multiport_mutex);
	3481	+
	3482	+ spin_lock(&ibdev->port[port_num].mp.mpi_lock);
	3483	+ if (ibdev->port[port_num].mp.mpi) {
	3484	+ mlx5_ib_dbg(ibdev, "port %d already affiliated.\n",
	3485	+ port_num + 1);
	3486	+ spin_unlock(&ibdev->port[port_num].mp.mpi_lock);
	3487	+ return false;
	3488	+ }
	3489	+
	3490	+ ibdev->port[port_num].mp.mpi = mpi;
	3491	+ mpi->ibdev = ibdev;
	3492	+ mpi->mdev_events.notifier_call = NULL;
	3493	+ spin_unlock(&ibdev->port[port_num].mp.mpi_lock);
	3494	+
	3495	+ err = mlx5_nic_vport_affiliate_multiport(ibdev->mdev, mpi->mdev);
	3496	+ if (err)
	3497	+ goto unbind;
	3498	+
	3499	+ err = get_port_caps(ibdev, mlx5_core_native_port_num(mpi->mdev));
	3500	+ if (err)
	3501	+ goto unbind;
	3502	+
	3503	+ err = mlx5_add_netdev_notifier(ibdev, port_num);
	3504	+ if (err) {
	3505	+ mlx5_ib_err(ibdev, "failed adding netdev notifier for port %u\n",
	3506	+ port_num + 1);
	3507	+ goto unbind;
	3508	+ }
	3509	+
	3510	+ mpi->mdev_events.notifier_call = mlx5_ib_event_slave_port;
	3511	+ mlx5_notifier_register(mpi->mdev, &mpi->mdev_events);
	3512	+
	3513	+ mlx5_ib_init_cong_debugfs(ibdev, port_num);
	3514	+
	3515	+ return true;
	3516	+
	3517	+unbind:
	3518	+ mlx5_ib_unbind_slave_port(ibdev, mpi);
	3519	+ return false;
	3520	+}
	3521	+
	3522	+static int mlx5_ib_init_multiport_master(struct mlx5_ib_dev *dev)
	3523	+{
	3524	+ int port_num = mlx5_core_native_port_num(dev->mdev) - 1;
	3525	+ enum rdma_link_layer ll = mlx5_ib_port_link_layer(&dev->ib_dev,
	3526	+ port_num + 1);
	3527	+ struct mlx5_ib_multiport_info *mpi;
	3528	+ int err;
	3529	+ int i;
	3530	+
	3531	+ if (!mlx5_core_is_mp_master(dev->mdev) \|\| ll != IB_LINK_LAYER_ETHERNET)
	3532	+ return 0;
	3533	+
	3534	+ err = mlx5_query_nic_vport_system_image_guid(dev->mdev,
	3535	+ &dev->sys_image_guid);
	3536	+ if (err)
	3537	+ return err;
	3538	+
	3539	+ err = mlx5_nic_vport_enable_roce(dev->mdev);
	3540	+ if (err)
	3541	+ return err;
	3542	+
	3543	+ mutex_lock(&mlx5_ib_multiport_mutex);
	3544	+ for (i = 0; i < dev->num_ports; i++) {
	3545	+ bool bound = false;
	3546	+
	3547	+ /* build a stub multiport info struct for the native port. */
	3548	+ if (i == port_num) {
	3549	+ mpi = kzalloc(sizeof(*mpi), GFP_KERNEL);
	3550	+ if (!mpi) {
	3551	+ mutex_unlock(&mlx5_ib_multiport_mutex);
	3552	+ mlx5_nic_vport_disable_roce(dev->mdev);
	3553	+ return -ENOMEM;
	3554	+ }
	3555	+
	3556	+ mpi->is_master = true;
	3557	+ mpi->mdev = dev->mdev;
	3558	+ mpi->sys_image_guid = dev->sys_image_guid;
	3559	+ dev->port[i].mp.mpi = mpi;
	3560	+ mpi->ibdev = dev;
	3561	+ mpi = NULL;
	3562	+ continue;
	3563	+ }
	3564	+
	3565	+ list_for_each_entry(mpi, &mlx5_ib_unaffiliated_port_list,
	3566	+ list) {
	3567	+ if (dev->sys_image_guid == mpi->sys_image_guid &&
	3568	+ (mlx5_core_native_port_num(mpi->mdev) - 1) == i) {
	3569	+ bound = mlx5_ib_bind_slave_port(dev, mpi);
	3570	+ }
	3571	+
	3572	+ if (bound) {
	3573	+ dev_dbg(mpi->mdev->device,
	3574	+ "removing port from unaffiliated list.\n");
	3575	+ mlx5_ib_dbg(dev, "port %d bound\n", i + 1);
	3576	+ list_del(&mpi->list);
	3577	+ break;
	3578	+ }
	3579	+ }
	3580	+ if (!bound) {
	3581	+ get_port_caps(dev, i + 1);
	3582	+ mlx5_ib_dbg(dev, "no free port found for port %d\n",
	3583	+ i + 1);
	3584	+ }
	3585	+ }
	3586	+
	3587	+ list_add_tail(&dev->ib_dev_list, &mlx5_ib_dev_list);
	3588	+ mutex_unlock(&mlx5_ib_multiport_mutex);
	3589	+ return err;
	3590	+}
	3591	+
	3592	+static void mlx5_ib_cleanup_multiport_master(struct mlx5_ib_dev *dev)
	3593	+{
	3594	+ int port_num = mlx5_core_native_port_num(dev->mdev) - 1;
	3595	+ enum rdma_link_layer ll = mlx5_ib_port_link_layer(&dev->ib_dev,
	3596	+ port_num + 1);
	3597	+ int i;
	3598	+
	3599	+ if (!mlx5_core_is_mp_master(dev->mdev) \|\| ll != IB_LINK_LAYER_ETHERNET)
	3600	+ return;
	3601	+
	3602	+ mutex_lock(&mlx5_ib_multiport_mutex);
	3603	+ for (i = 0; i < dev->num_ports; i++) {
	3604	+ if (dev->port[i].mp.mpi) {
	3605	+ /* Destroy the native port stub */
	3606	+ if (i == port_num) {
	3607	+ kfree(dev->port[i].mp.mpi);
	3608	+ dev->port[i].mp.mpi = NULL;
	3609	+ } else {
	3610	+ mlx5_ib_dbg(dev, "unbinding port_num: %d\n", i + 1);
	3611	+ list_add_tail(&dev->port[i].mp.mpi->list,
	3612	+ &mlx5_ib_unaffiliated_port_list);
	3613	+ mlx5_ib_unbind_slave_port(dev, dev->port[i].mp.mpi);
	3614	+ }
	3615	+ }
	3616	+ }
	3617	+
	3618	+ mlx5_ib_dbg(dev, "removing from devlist\n");
	3619	+ list_del(&dev->ib_dev_list);
	3620	+ mutex_unlock(&mlx5_ib_multiport_mutex);
	3621	+
	3622	+ mlx5_nic_vport_disable_roce(dev->mdev);
	3623	+}
	3624	+
	3625	+static int mmap_obj_cleanup(struct ib_uobject *uobject,
	3626	+ enum rdma_remove_reason why,
	3627	+ struct uverbs_attr_bundle *attrs)
	3628	+{
	3629	+ struct mlx5_user_mmap_entry *obj = uobject->object;
	3630	+
	3631	+ rdma_user_mmap_entry_remove(&obj->rdma_entry);
	3632	+ return 0;
	3633	+}
	3634	+
	3635	+static int mlx5_rdma_user_mmap_entry_insert(struct mlx5_ib_ucontext *c,
	3636	+ struct mlx5_user_mmap_entry *entry,
	3637	+ size_t length)
	3638	+{
	3639	+ return rdma_user_mmap_entry_insert_range(
	3640	+ &c->ibucontext, &entry->rdma_entry, length,
	3641	+ (MLX5_IB_MMAP_OFFSET_START << 16),
	3642	+ ((MLX5_IB_MMAP_OFFSET_END << 16) + (1UL << 16) - 1));
	3643	+}
	3644	+
	3645	+static struct mlx5_user_mmap_entry *
	3646	+alloc_var_entry(struct mlx5_ib_ucontext *c)
	3647	+{
	3648	+ struct mlx5_user_mmap_entry *entry;
	3649	+ struct mlx5_var_table *var_table;
	3650	+ u32 page_idx;
	3651	+ int err;
	3652	+
	3653	+ var_table = &to_mdev(c->ibucontext.device)->var_table;
	3654	+ entry = kzalloc(sizeof(*entry), GFP_KERNEL);
	3655	+ if (!entry)
	3656	+ return ERR_PTR(-ENOMEM);
	3657	+
	3658	+ mutex_lock(&var_table->bitmap_lock);
	3659	+ page_idx = find_first_zero_bit(var_table->bitmap,
	3660	+ var_table->num_var_hw_entries);
	3661	+ if (page_idx >= var_table->num_var_hw_entries) {
	3662	+ err = -ENOSPC;
	3663	+ mutex_unlock(&var_table->bitmap_lock);
	3664	+ goto end;
	3665	+ }
	3666	+
	3667	+ set_bit(page_idx, var_table->bitmap);
	3668	+ mutex_unlock(&var_table->bitmap_lock);
	3669	+
	3670	+ entry->address = var_table->hw_start_addr +
	3671	+ (page_idx * var_table->stride_size);
	3672	+ entry->page_idx = page_idx;
	3673	+ entry->mmap_flag = MLX5_IB_MMAP_TYPE_VAR;
	3674	+
	3675	+ err = mlx5_rdma_user_mmap_entry_insert(c, entry,
	3676	+ var_table->stride_size);
	3677	+ if (err)
	3678	+ goto err_insert;
	3679	+
	3680	+ return entry;
	3681	+
	3682	+err_insert:
	3683	+ mutex_lock(&var_table->bitmap_lock);
	3684	+ clear_bit(page_idx, var_table->bitmap);
	3685	+ mutex_unlock(&var_table->bitmap_lock);
	3686	+end:
	3687	+ kfree(entry);
	3688	+ return ERR_PTR(err);
	3689	+}
	3690	+
	3691	+static int UVERBS_HANDLER(MLX5_IB_METHOD_VAR_OBJ_ALLOC)(
	3692	+ struct uverbs_attr_bundle *attrs)
	3693	+{
	3694	+ struct ib_uobject *uobj = uverbs_attr_get_uobject(
	3695	+ attrs, MLX5_IB_ATTR_VAR_OBJ_ALLOC_HANDLE);
	3696	+ struct mlx5_ib_ucontext *c;
	3697	+ struct mlx5_user_mmap_entry *entry;
	3698	+ u64 mmap_offset;
	3699	+ u32 length;
	3700	+ int err;
	3701	+
	3702	+ c = to_mucontext(ib_uverbs_get_ucontext(attrs));
	3703	+ if (IS_ERR(c))
	3704	+ return PTR_ERR(c);
	3705	+
	3706	+ entry = alloc_var_entry(c);
	3707	+ if (IS_ERR(entry))
	3708	+ return PTR_ERR(entry);
	3709	+
	3710	+ mmap_offset = mlx5_entry_to_mmap_offset(entry);
	3711	+ length = entry->rdma_entry.npages * PAGE_SIZE;
	3712	+ uobj->object = entry;
	3713	+ uverbs_finalize_uobj_create(attrs, MLX5_IB_ATTR_VAR_OBJ_ALLOC_HANDLE);
	3714	+
	3715	+ err = uverbs_copy_to(attrs, MLX5_IB_ATTR_VAR_OBJ_ALLOC_MMAP_OFFSET,
	3716	+ &mmap_offset, sizeof(mmap_offset));
	3717	+ if (err)
	3718	+ return err;
	3719	+
	3720	+ err = uverbs_copy_to(attrs, MLX5_IB_ATTR_VAR_OBJ_ALLOC_PAGE_ID,
	3721	+ &entry->page_idx, sizeof(entry->page_idx));
	3722	+ if (err)
	3723	+ return err;
	3724	+
	3725	+ err = uverbs_copy_to(attrs, MLX5_IB_ATTR_VAR_OBJ_ALLOC_MMAP_LENGTH,
	3726	+ &length, sizeof(length));
	3727	+ return err;
	3728	+}
	3729	+
	3730	+DECLARE_UVERBS_NAMED_METHOD(
	3731	+ MLX5_IB_METHOD_VAR_OBJ_ALLOC,
	3732	+ UVERBS_ATTR_IDR(MLX5_IB_ATTR_VAR_OBJ_ALLOC_HANDLE,
	3733	+ MLX5_IB_OBJECT_VAR,
	3734	+ UVERBS_ACCESS_NEW,
	3735	+ UA_MANDATORY),
	3736	+ UVERBS_ATTR_PTR_OUT(MLX5_IB_ATTR_VAR_OBJ_ALLOC_PAGE_ID,
	3737	+ UVERBS_ATTR_TYPE(u32),
	3738	+ UA_MANDATORY),
	3739	+ UVERBS_ATTR_PTR_OUT(MLX5_IB_ATTR_VAR_OBJ_ALLOC_MMAP_LENGTH,
	3740	+ UVERBS_ATTR_TYPE(u32),
	3741	+ UA_MANDATORY),
	3742	+ UVERBS_ATTR_PTR_OUT(MLX5_IB_ATTR_VAR_OBJ_ALLOC_MMAP_OFFSET,
	3743	+ UVERBS_ATTR_TYPE(u64),
	3744	+ UA_MANDATORY));
	3745	+
	3746	+DECLARE_UVERBS_NAMED_METHOD_DESTROY(
	3747	+ MLX5_IB_METHOD_VAR_OBJ_DESTROY,
	3748	+ UVERBS_ATTR_IDR(MLX5_IB_ATTR_VAR_OBJ_DESTROY_HANDLE,
	3749	+ MLX5_IB_OBJECT_VAR,
	3750	+ UVERBS_ACCESS_DESTROY,
	3751	+ UA_MANDATORY));
	3752	+
	3753	+DECLARE_UVERBS_NAMED_OBJECT(MLX5_IB_OBJECT_VAR,
	3754	+ UVERBS_TYPE_ALLOC_IDR(mmap_obj_cleanup),
	3755	+ &UVERBS_METHOD(MLX5_IB_METHOD_VAR_OBJ_ALLOC),
	3756	+ &UVERBS_METHOD(MLX5_IB_METHOD_VAR_OBJ_DESTROY));
	3757	+
	3758	+static bool var_is_supported(struct ib_device *device)
	3759	+{
	3760	+ struct mlx5_ib_dev *dev = to_mdev(device);
	3761	+
	3762	+ return (MLX5_CAP_GEN_64(dev->mdev, general_obj_types) &
	3763	+ MLX5_GENERAL_OBJ_TYPES_CAP_VIRTIO_NET_Q);
	3764	+}
	3765	+
	3766	+static struct mlx5_user_mmap_entry *
	3767	+alloc_uar_entry(struct mlx5_ib_ucontext *c,
	3768	+ enum mlx5_ib_uapi_uar_alloc_type alloc_type)
	3769	+{
	3770	+ struct mlx5_user_mmap_entry *entry;
	3771	+ struct mlx5_ib_dev *dev;
	3772	+ u32 uar_index;
	3773	+ int err;
	3774	+
	3775	+ entry = kzalloc(sizeof(*entry), GFP_KERNEL);
	3776	+ if (!entry)
	3777	+ return ERR_PTR(-ENOMEM);
	3778	+
	3779	+ dev = to_mdev(c->ibucontext.device);
	3780	+ err = mlx5_cmd_alloc_uar(dev->mdev, &uar_index);
	3781	+ if (err)
	3782	+ goto end;
	3783	+
	3784	+ entry->page_idx = uar_index;
	3785	+ entry->address = uar_index2paddress(dev, uar_index);
	3786	+ if (alloc_type == MLX5_IB_UAPI_UAR_ALLOC_TYPE_BF)
	3787	+ entry->mmap_flag = MLX5_IB_MMAP_TYPE_UAR_WC;
	3788	+ else
	3789	+ entry->mmap_flag = MLX5_IB_MMAP_TYPE_UAR_NC;
	3790	+
	3791	+ err = mlx5_rdma_user_mmap_entry_insert(c, entry, PAGE_SIZE);
	3792	+ if (err)
	3793	+ goto err_insert;
	3794	+
	3795	+ return entry;
	3796	+
	3797	+err_insert:
	3798	+ mlx5_cmd_free_uar(dev->mdev, uar_index);
	3799	+end:
	3800	+ kfree(entry);
	3801	+ return ERR_PTR(err);
	3802	+}
	3803	+
	3804	+static int UVERBS_HANDLER(MLX5_IB_METHOD_UAR_OBJ_ALLOC)(
	3805	+ struct uverbs_attr_bundle *attrs)
	3806	+{
	3807	+ struct ib_uobject *uobj = uverbs_attr_get_uobject(
	3808	+ attrs, MLX5_IB_ATTR_UAR_OBJ_ALLOC_HANDLE);
	3809	+ enum mlx5_ib_uapi_uar_alloc_type alloc_type;
	3810	+ struct mlx5_ib_ucontext *c;
	3811	+ struct mlx5_user_mmap_entry *entry;
	3812	+ u64 mmap_offset;
	3813	+ u32 length;
	3814	+ int err;
	3815	+
	3816	+ c = to_mucontext(ib_uverbs_get_ucontext(attrs));
	3817	+ if (IS_ERR(c))
	3818	+ return PTR_ERR(c);
	3819	+
	3820	+ err = uverbs_get_const(&alloc_type, attrs,
	3821	+ MLX5_IB_ATTR_UAR_OBJ_ALLOC_TYPE);
	3822	+ if (err)
	3823	+ return err;
	3824	+
	3825	+ if (alloc_type != MLX5_IB_UAPI_UAR_ALLOC_TYPE_BF &&
	3826	+ alloc_type != MLX5_IB_UAPI_UAR_ALLOC_TYPE_NC)
	3827	+ return -EOPNOTSUPP;
	3828	+
	3829	+ if (!to_mdev(c->ibucontext.device)->wc_support &&
	3830	+ alloc_type == MLX5_IB_UAPI_UAR_ALLOC_TYPE_BF)
	3831	+ return -EOPNOTSUPP;
	3832	+
	3833	+ entry = alloc_uar_entry(c, alloc_type);
	3834	+ if (IS_ERR(entry))
	3835	+ return PTR_ERR(entry);
	3836	+
	3837	+ mmap_offset = mlx5_entry_to_mmap_offset(entry);
	3838	+ length = entry->rdma_entry.npages * PAGE_SIZE;
	3839	+ uobj->object = entry;
	3840	+ uverbs_finalize_uobj_create(attrs, MLX5_IB_ATTR_UAR_OBJ_ALLOC_HANDLE);
	3841	+
	3842	+ err = uverbs_copy_to(attrs, MLX5_IB_ATTR_UAR_OBJ_ALLOC_MMAP_OFFSET,
	3843	+ &mmap_offset, sizeof(mmap_offset));
	3844	+ if (err)
	3845	+ return err;
	3846	+
	3847	+ err = uverbs_copy_to(attrs, MLX5_IB_ATTR_UAR_OBJ_ALLOC_PAGE_ID,
	3848	+ &entry->page_idx, sizeof(entry->page_idx));
	3849	+ if (err)
	3850	+ return err;
	3851	+
	3852	+ err = uverbs_copy_to(attrs, MLX5_IB_ATTR_UAR_OBJ_ALLOC_MMAP_LENGTH,
	3853	+ &length, sizeof(length));
	3854	+ return err;
	3855	+}
	3856	+
	3857	+DECLARE_UVERBS_NAMED_METHOD(
	3858	+ MLX5_IB_METHOD_UAR_OBJ_ALLOC,
	3859	+ UVERBS_ATTR_IDR(MLX5_IB_ATTR_UAR_OBJ_ALLOC_HANDLE,
	3860	+ MLX5_IB_OBJECT_UAR,
	3861	+ UVERBS_ACCESS_NEW,
	3862	+ UA_MANDATORY),
	3863	+ UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_UAR_OBJ_ALLOC_TYPE,
	3864	+ enum mlx5_ib_uapi_uar_alloc_type,
	3865	+ UA_MANDATORY),
	3866	+ UVERBS_ATTR_PTR_OUT(MLX5_IB_ATTR_UAR_OBJ_ALLOC_PAGE_ID,
	3867	+ UVERBS_ATTR_TYPE(u32),
	3868	+ UA_MANDATORY),
	3869	+ UVERBS_ATTR_PTR_OUT(MLX5_IB_ATTR_UAR_OBJ_ALLOC_MMAP_LENGTH,
	3870	+ UVERBS_ATTR_TYPE(u32),
	3871	+ UA_MANDATORY),
	3872	+ UVERBS_ATTR_PTR_OUT(MLX5_IB_ATTR_UAR_OBJ_ALLOC_MMAP_OFFSET,
	3873	+ UVERBS_ATTR_TYPE(u64),
	3874	+ UA_MANDATORY));
	3875	+
	3876	+DECLARE_UVERBS_NAMED_METHOD_DESTROY(
	3877	+ MLX5_IB_METHOD_UAR_OBJ_DESTROY,
	3878	+ UVERBS_ATTR_IDR(MLX5_IB_ATTR_UAR_OBJ_DESTROY_HANDLE,
	3879	+ MLX5_IB_OBJECT_UAR,
	3880	+ UVERBS_ACCESS_DESTROY,
	3881	+ UA_MANDATORY));
	3882	+
	3883	+DECLARE_UVERBS_NAMED_OBJECT(MLX5_IB_OBJECT_UAR,
	3884	+ UVERBS_TYPE_ALLOC_IDR(mmap_obj_cleanup),
	3885	+ &UVERBS_METHOD(MLX5_IB_METHOD_UAR_OBJ_ALLOC),
	3886	+ &UVERBS_METHOD(MLX5_IB_METHOD_UAR_OBJ_DESTROY));
	3887	+
	3888	+ADD_UVERBS_ATTRIBUTES_SIMPLE(
	3889	+ mlx5_ib_dm,
	3890	+ UVERBS_OBJECT_DM,
	3891	+ UVERBS_METHOD_DM_ALLOC,
	3892	+ UVERBS_ATTR_PTR_OUT(MLX5_IB_ATTR_ALLOC_DM_RESP_START_OFFSET,
	3893	+ UVERBS_ATTR_TYPE(u64),
	3894	+ UA_MANDATORY),
	3895	+ UVERBS_ATTR_PTR_OUT(MLX5_IB_ATTR_ALLOC_DM_RESP_PAGE_INDEX,
	3896	+ UVERBS_ATTR_TYPE(u16),
	3897	+ UA_OPTIONAL),
	3898	+ UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_ALLOC_DM_REQ_TYPE,
	3899	+ enum mlx5_ib_uapi_dm_type,
	3900	+ UA_OPTIONAL));
	3901	+
	3902	+ADD_UVERBS_ATTRIBUTES_SIMPLE(
	3903	+ mlx5_ib_flow_action,
	3904	+ UVERBS_OBJECT_FLOW_ACTION,
	3905	+ UVERBS_METHOD_FLOW_ACTION_ESP_CREATE,
	3906	+ UVERBS_ATTR_FLAGS_IN(MLX5_IB_ATTR_CREATE_FLOW_ACTION_FLAGS,
	3907	+ enum mlx5_ib_uapi_flow_action_flags));
	3908	+
	3909	+ADD_UVERBS_ATTRIBUTES_SIMPLE(
	3910	+ mlx5_ib_query_context,
	3911	+ UVERBS_OBJECT_DEVICE,
	3912	+ UVERBS_METHOD_QUERY_CONTEXT,
	3913	+ UVERBS_ATTR_PTR_OUT(
	3914	+ MLX5_IB_ATTR_QUERY_CONTEXT_RESP_UCTX,
	3915	+ UVERBS_ATTR_STRUCT(struct mlx5_ib_alloc_ucontext_resp,
	3916	+ dump_fill_mkey),
	3917	+ UA_MANDATORY));
	3918	+
	3919	+static const struct uapi_definition mlx5_ib_defs[] = {
	3920	+ UAPI_DEF_CHAIN(mlx5_ib_devx_defs),
	3921	+ UAPI_DEF_CHAIN(mlx5_ib_flow_defs),
	3922	+ UAPI_DEF_CHAIN(mlx5_ib_qos_defs),
	3923	+ UAPI_DEF_CHAIN(mlx5_ib_std_types_defs),
	3924	+
	3925	+ UAPI_DEF_CHAIN_OBJ_TREE(UVERBS_OBJECT_FLOW_ACTION,
	3926	+ &mlx5_ib_flow_action),
	3927	+ UAPI_DEF_CHAIN_OBJ_TREE(UVERBS_OBJECT_DM, &mlx5_ib_dm),
	3928	+ UAPI_DEF_CHAIN_OBJ_TREE(UVERBS_OBJECT_DEVICE, &mlx5_ib_query_context),
	3929	+ UAPI_DEF_CHAIN_OBJ_TREE_NAMED(MLX5_IB_OBJECT_VAR,
	3930	+ UAPI_DEF_IS_OBJ_SUPPORTED(var_is_supported)),
	3931	+ UAPI_DEF_CHAIN_OBJ_TREE_NAMED(MLX5_IB_OBJECT_UAR),
	3932	+ {}
	3933	+};
	3934	+
	3935	+static void mlx5_ib_stage_init_cleanup(struct mlx5_ib_dev *dev)
	3936	+{
	3937	+ mlx5_ib_cleanup_multiport_master(dev);
	3938	+ WARN_ON(!xa_empty(&dev->odp_mkeys));
	3939	+ cleanup_srcu_struct(&dev->odp_srcu);
	3940	+ mutex_destroy(&dev->cap_mask_mutex);
	3941	+ WARN_ON(!xa_empty(&dev->sig_mrs));
	3942	+ WARN_ON(!bitmap_empty(dev->dm.memic_alloc_pages, MLX5_MAX_MEMIC_PAGES));
	3943	+}
	3944	+
	3945	+static int mlx5_ib_stage_init_init(struct mlx5_ib_dev *dev)
	3946	+{
	3947	+ struct mlx5_core_dev *mdev = dev->mdev;
	3948	+ int err;
	3949	+ int i;
	3950	+
	3951	+ for (i = 0; i < dev->num_ports; i++) {
	3952	+ spin_lock_init(&dev->port[i].mp.mpi_lock);
	3953	+ rwlock_init(&dev->port[i].roce.netdev_lock);
	3954	+ dev->port[i].roce.dev = dev;
	3955	+ dev->port[i].roce.native_port_num = i + 1;
	3956	+ dev->port[i].roce.last_port_state = IB_PORT_DOWN;
	3957	+ }
	3958	+
	3959	+ mlx5_ib_internal_fill_odp_caps(dev);
	3960	+
	3961	+ err = mlx5_ib_init_multiport_master(dev);
	3962	+ if (err)
	3963	+ return err;
	3964	+
	3965	+ err = set_has_smi_cap(dev);
	3966	+ if (err)
	3967	+ goto err_mp;
	3968	+
	3969	+ if (!mlx5_core_mp_enabled(mdev)) {
	3970	+ for (i = 1; i <= dev->num_ports; i++) {
	3971	+ err = get_port_caps(dev, i);
	3972	+ if (err)
	3973	+ break;
	3974	+ }
	3975	+ } else {
	3976	+ err = get_port_caps(dev, mlx5_core_native_port_num(mdev));
	3977	+ }
	3978	+ if (err)
	3979	+ goto err_mp;
	3980	+
	3981	+ if (mlx5_use_mad_ifc(dev))
	3982	+ get_ext_port_caps(dev);
	3983	+
	3984	+ dev->ib_dev.node_type = RDMA_NODE_IB_CA;
	3985	+ dev->ib_dev.local_dma_lkey = 0 /* not supported for now */;
	3986	+ dev->ib_dev.phys_port_cnt = dev->num_ports;
	3987	+ dev->ib_dev.num_comp_vectors = mlx5_comp_vectors_count(mdev);
	3988	+ dev->ib_dev.dev.parent = mdev->device;
	3989	+ dev->ib_dev.lag_flags = RDMA_LAG_FLAGS_HASH_ALL_SLAVES;
	3990	+
	3991	+ err = init_srcu_struct(&dev->odp_srcu);
	3992	+ if (err)
	3993	+ goto err_mp;
	3994	+
	3995	+ mutex_init(&dev->cap_mask_mutex);
	3996	+ INIT_LIST_HEAD(&dev->qp_list);
	3997	+ spin_lock_init(&dev->reset_flow_resource_lock);
	3998	+ xa_init(&dev->odp_mkeys);
	3999	+ xa_init(&dev->sig_mrs);
	4000	+ atomic_set(&dev->mkey_var, 0);
	4001	+
	4002	+ spin_lock_init(&dev->dm.lock);
	4003	+ dev->dm.dev = mdev;
	4004	+ return 0;
	4005	+
	4006	+err_mp:
	4007	+ mlx5_ib_cleanup_multiport_master(dev);
	4008	+ return err;
	4009	+}
	4010	+
	4011	+static int mlx5_ib_enable_driver(struct ib_device *dev)
	4012	+{
	4013	+ struct mlx5_ib_dev *mdev = to_mdev(dev);
	4014	+ int ret;
	4015	+
	4016	+ ret = mlx5_ib_test_wc(mdev);
	4017	+ mlx5_ib_dbg(mdev, "Write-Combining %s",
	4018	+ mdev->wc_support ? "supported" : "not supported");
	4019	+
	4020	+ return ret;
	4021	+}
	4022	+
	4023	+static const struct ib_device_ops mlx5_ib_dev_ops = {
	4024	+ .owner = THIS_MODULE,
	4025	+ .driver_id = RDMA_DRIVER_MLX5,
	4026	+ .uverbs_abi_ver = MLX5_IB_UVERBS_ABI_VERSION,
	4027	+
	4028	+ .add_gid = mlx5_ib_add_gid,
	4029	+ .alloc_mr = mlx5_ib_alloc_mr,
	4030	+ .alloc_mr_integrity = mlx5_ib_alloc_mr_integrity,
	4031	+ .alloc_pd = mlx5_ib_alloc_pd,
	4032	+ .alloc_ucontext = mlx5_ib_alloc_ucontext,
	4033	+ .attach_mcast = mlx5_ib_mcg_attach,
	4034	+ .check_mr_status = mlx5_ib_check_mr_status,
	4035	+ .create_ah = mlx5_ib_create_ah,
	4036	+ .create_cq = mlx5_ib_create_cq,
	4037	+ .create_qp = mlx5_ib_create_qp,
	4038	+ .create_srq = mlx5_ib_create_srq,
	4039	+ .dealloc_pd = mlx5_ib_dealloc_pd,
	4040	+ .dealloc_ucontext = mlx5_ib_dealloc_ucontext,
	4041	+ .del_gid = mlx5_ib_del_gid,
	4042	+ .dereg_mr = mlx5_ib_dereg_mr,
	4043	+ .destroy_ah = mlx5_ib_destroy_ah,
	4044	+ .destroy_cq = mlx5_ib_destroy_cq,
	4045	+ .destroy_qp = mlx5_ib_destroy_qp,
	4046	+ .destroy_srq = mlx5_ib_destroy_srq,
	4047	+ .detach_mcast = mlx5_ib_mcg_detach,
	4048	+ .disassociate_ucontext = mlx5_ib_disassociate_ucontext,
	4049	+ .drain_rq = mlx5_ib_drain_rq,
	4050	+ .drain_sq = mlx5_ib_drain_sq,
	4051	+ .enable_driver = mlx5_ib_enable_driver,
	4052	+ .get_dev_fw_str = get_dev_fw_str,
	4053	+ .get_dma_mr = mlx5_ib_get_dma_mr,
	4054	+ .get_link_layer = mlx5_ib_port_link_layer,
	4055	+ .map_mr_sg = mlx5_ib_map_mr_sg,
	4056	+ .map_mr_sg_pi = mlx5_ib_map_mr_sg_pi,
	4057	+ .mmap = mlx5_ib_mmap,
	4058	+ .mmap_free = mlx5_ib_mmap_free,
	4059	+ .modify_cq = mlx5_ib_modify_cq,
	4060	+ .modify_device = mlx5_ib_modify_device,
	4061	+ .modify_port = mlx5_ib_modify_port,
	4062	+ .modify_qp = mlx5_ib_modify_qp,
	4063	+ .modify_srq = mlx5_ib_modify_srq,
	4064	+ .poll_cq = mlx5_ib_poll_cq,
	4065	+ .post_recv = mlx5_ib_post_recv_nodrain,
	4066	+ .post_send = mlx5_ib_post_send_nodrain,
	4067	+ .post_srq_recv = mlx5_ib_post_srq_recv,
	4068	+ .process_mad = mlx5_ib_process_mad,
	4069	+ .query_ah = mlx5_ib_query_ah,
	4070	+ .query_device = mlx5_ib_query_device,
	4071	+ .query_gid = mlx5_ib_query_gid,
	4072	+ .query_pkey = mlx5_ib_query_pkey,
	4073	+ .query_qp = mlx5_ib_query_qp,
	4074	+ .query_srq = mlx5_ib_query_srq,
	4075	+ .query_ucontext = mlx5_ib_query_ucontext,
	4076	+ .reg_user_mr = mlx5_ib_reg_user_mr,
	4077	+ .req_notify_cq = mlx5_ib_arm_cq,
	4078	+ .rereg_user_mr = mlx5_ib_rereg_user_mr,
	4079	+ .resize_cq = mlx5_ib_resize_cq,
	4080	+
	4081	+ INIT_RDMA_OBJ_SIZE(ib_ah, mlx5_ib_ah, ibah),
	4082	+ INIT_RDMA_OBJ_SIZE(ib_counters, mlx5_ib_mcounters, ibcntrs),
	4083	+ INIT_RDMA_OBJ_SIZE(ib_cq, mlx5_ib_cq, ibcq),
	4084	+ INIT_RDMA_OBJ_SIZE(ib_pd, mlx5_ib_pd, ibpd),
	4085	+ INIT_RDMA_OBJ_SIZE(ib_srq, mlx5_ib_srq, ibsrq),
	4086	+ INIT_RDMA_OBJ_SIZE(ib_ucontext, mlx5_ib_ucontext, ibucontext),
	4087	+};
	4088	+
	4089	+static const struct ib_device_ops mlx5_ib_dev_ipoib_enhanced_ops = {
	4090	+ .rdma_netdev_get_params = mlx5_ib_rn_get_params,
	4091	+};
	4092	+
	4093	+static const struct ib_device_ops mlx5_ib_dev_sriov_ops = {
	4094	+ .get_vf_config = mlx5_ib_get_vf_config,
	4095	+ .get_vf_guid = mlx5_ib_get_vf_guid,
	4096	+ .get_vf_stats = mlx5_ib_get_vf_stats,
	4097	+ .set_vf_guid = mlx5_ib_set_vf_guid,
	4098	+ .set_vf_link_state = mlx5_ib_set_vf_link_state,
	4099	+};
	4100	+
	4101	+static const struct ib_device_ops mlx5_ib_dev_mw_ops = {
	4102	+ .alloc_mw = mlx5_ib_alloc_mw,
	4103	+ .dealloc_mw = mlx5_ib_dealloc_mw,
	4104	+
	4105	+ INIT_RDMA_OBJ_SIZE(ib_mw, mlx5_ib_mw, ibmw),
	4106	+};
	4107	+
	4108	+static const struct ib_device_ops mlx5_ib_dev_xrc_ops = {
	4109	+ .alloc_xrcd = mlx5_ib_alloc_xrcd,
	4110	+ .dealloc_xrcd = mlx5_ib_dealloc_xrcd,
	4111	+
	4112	+ INIT_RDMA_OBJ_SIZE(ib_xrcd, mlx5_ib_xrcd, ibxrcd),
	4113	+};
	4114	+
	4115	+static const struct ib_device_ops mlx5_ib_dev_dm_ops = {
	4116	+ .alloc_dm = mlx5_ib_alloc_dm,
	4117	+ .dealloc_dm = mlx5_ib_dealloc_dm,
	4118	+ .reg_dm_mr = mlx5_ib_reg_dm_mr,
	4119	+};
	4120	+
	4121	+static int mlx5_ib_init_var_table(struct mlx5_ib_dev *dev)
	4122	+{
	4123	+ struct mlx5_core_dev *mdev = dev->mdev;
	4124	+ struct mlx5_var_table *var_table = &dev->var_table;
	4125	+ u8 log_doorbell_bar_size;
	4126	+ u8 log_doorbell_stride;
	4127	+ u64 bar_size;
	4128	+
	4129	+ log_doorbell_bar_size = MLX5_CAP_DEV_VDPA_EMULATION(mdev,
	4130	+ log_doorbell_bar_size);
	4131	+ log_doorbell_stride = MLX5_CAP_DEV_VDPA_EMULATION(mdev,
	4132	+ log_doorbell_stride);
	4133	+ var_table->hw_start_addr = dev->mdev->bar_addr +
	4134	+ MLX5_CAP64_DEV_VDPA_EMULATION(mdev,
	4135	+ doorbell_bar_offset);
	4136	+ bar_size = (1ULL << log_doorbell_bar_size) * 4096;
	4137	+ var_table->stride_size = 1ULL << log_doorbell_stride;
	4138	+ var_table->num_var_hw_entries = div_u64(bar_size,
	4139	+ var_table->stride_size);
	4140	+ mutex_init(&var_table->bitmap_lock);
	4141	+ var_table->bitmap = bitmap_zalloc(var_table->num_var_hw_entries,
	4142	+ GFP_KERNEL);
	4143	+ return (var_table->bitmap) ? 0 : -ENOMEM;
	4144	+}
	4145	+
	4146	+static void mlx5_ib_stage_caps_cleanup(struct mlx5_ib_dev *dev)
	4147	+{
	4148	+ bitmap_free(dev->var_table.bitmap);
	4149	+}
	4150	+
	4151	+static int mlx5_ib_stage_caps_init(struct mlx5_ib_dev *dev)
	4152	+{
	4153	+ struct mlx5_core_dev *mdev = dev->mdev;
	4154	+ int err;
	4155	+
	4156	+ dev->ib_dev.uverbs_cmd_mask =
	4157	+ (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) \|
	4158	+ (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) \|
	4159	+ (1ull << IB_USER_VERBS_CMD_QUERY_PORT) \|
	4160	+ (1ull << IB_USER_VERBS_CMD_ALLOC_PD) \|
	4161	+ (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) \|
	4162	+ (1ull << IB_USER_VERBS_CMD_CREATE_AH) \|
	4163	+ (1ull << IB_USER_VERBS_CMD_DESTROY_AH) \|
	4164	+ (1ull << IB_USER_VERBS_CMD_REG_MR) \|
	4165	+ (1ull << IB_USER_VERBS_CMD_REREG_MR) \|
	4166	+ (1ull << IB_USER_VERBS_CMD_DEREG_MR) \|
	4167	+ (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) \|
	4168	+ (1ull << IB_USER_VERBS_CMD_CREATE_CQ) \|
	4169	+ (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) \|
	4170	+ (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) \|
	4171	+ (1ull << IB_USER_VERBS_CMD_CREATE_QP) \|
	4172	+ (1ull << IB_USER_VERBS_CMD_MODIFY_QP) \|
	4173	+ (1ull << IB_USER_VERBS_CMD_QUERY_QP) \|
	4174	+ (1ull << IB_USER_VERBS_CMD_DESTROY_QP) \|
	4175	+ (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) \|
	4176	+ (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) \|
	4177	+ (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) \|
	4178	+ (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) \|
	4179	+ (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) \|
	4180	+ (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) \|
	4181	+ (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ) \|
	4182	+ (1ull << IB_USER_VERBS_CMD_OPEN_QP);
	4183	+ dev->ib_dev.uverbs_ex_cmd_mask \|=
	4184	+ (1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) \|
	4185	+ (1ull << IB_USER_VERBS_EX_CMD_CREATE_QP) \|
	4186	+ (1ull << IB_USER_VERBS_EX_CMD_MODIFY_QP);
	4187	+
	4188	+ if (MLX5_CAP_GEN(mdev, ipoib_enhanced_offloads) &&
	4189	+ IS_ENABLED(CONFIG_MLX5_CORE_IPOIB))
	4190	+ ib_set_device_ops(&dev->ib_dev,
	4191	+ &mlx5_ib_dev_ipoib_enhanced_ops);
	4192	+
	4193	+ if (mlx5_core_is_pf(mdev))
	4194	+ ib_set_device_ops(&dev->ib_dev, &mlx5_ib_dev_sriov_ops);
	4195	+
	4196	+ dev->umr_fence = mlx5_get_umr_fence(MLX5_CAP_GEN(mdev, umr_fence));
	4197	+
	4198	+ if (MLX5_CAP_GEN(mdev, imaicl)) {
	4199	+ dev->ib_dev.uverbs_cmd_mask \|=
	4200	+ (1ull << IB_USER_VERBS_CMD_ALLOC_MW) \|
	4201	+ (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
	4202	+ ib_set_device_ops(&dev->ib_dev, &mlx5_ib_dev_mw_ops);
	4203	+ }
	4204	+
	4205	+ if (MLX5_CAP_GEN(mdev, xrc)) {
	4206	+ dev->ib_dev.uverbs_cmd_mask \|=
	4207	+ (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) \|
	4208	+ (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
	4209	+ ib_set_device_ops(&dev->ib_dev, &mlx5_ib_dev_xrc_ops);
	4210	+ }
	4211	+
	4212	+ if (MLX5_CAP_DEV_MEM(mdev, memic) \|\|
	4213	+ MLX5_CAP_GEN_64(dev->mdev, general_obj_types) &
	4214	+ MLX5_GENERAL_OBJ_TYPES_CAP_SW_ICM)
	4215	+ ib_set_device_ops(&dev->ib_dev, &mlx5_ib_dev_dm_ops);
	4216	+
	4217	+ ib_set_device_ops(&dev->ib_dev, &mlx5_ib_dev_ops);
	4218	+
	4219	+ if (IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS))
	4220	+ dev->ib_dev.driver_def = mlx5_ib_defs;
	4221	+
	4222	+ err = init_node_data(dev);
	4223	+ if (err)
	4224	+ return err;
	4225	+
	4226	+ if ((MLX5_CAP_GEN(dev->mdev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
	4227	+ (MLX5_CAP_GEN(dev->mdev, disable_local_lb_uc) \|\|
	4228	+ MLX5_CAP_GEN(dev->mdev, disable_local_lb_mc)))
	4229	+ mutex_init(&dev->lb.mutex);
	4230	+
	4231	+ if (MLX5_CAP_GEN_64(dev->mdev, general_obj_types) &
	4232	+ MLX5_GENERAL_OBJ_TYPES_CAP_VIRTIO_NET_Q) {
	4233	+ err = mlx5_ib_init_var_table(dev);
	4234	+ if (err)
	4235	+ return err;
	4236	+ }
	4237	+
	4238	+ dev->ib_dev.use_cq_dim = true;
	4239	+
	4240	+ return 0;
	4241	+}
	4242	+
	4243	+static const struct ib_device_ops mlx5_ib_dev_port_ops = {
	4244	+ .get_port_immutable = mlx5_port_immutable,
	4245	+ .query_port = mlx5_ib_query_port,
	4246	+};
	4247	+
	4248	+static int mlx5_ib_stage_non_default_cb(struct mlx5_ib_dev *dev)
	4249	+{
	4250	+ ib_set_device_ops(&dev->ib_dev, &mlx5_ib_dev_port_ops);
	4251	+ return 0;
	4252	+}
	4253	+
	4254	+static const struct ib_device_ops mlx5_ib_dev_port_rep_ops = {
	4255	+ .get_port_immutable = mlx5_port_rep_immutable,
	4256	+ .query_port = mlx5_ib_rep_query_port,
	4257	+};
	4258	+
	4259	+static int mlx5_ib_stage_raw_eth_non_default_cb(struct mlx5_ib_dev *dev)
	4260	+{
	4261	+ ib_set_device_ops(&dev->ib_dev, &mlx5_ib_dev_port_rep_ops);
	4262	+ return 0;
	4263	+}
	4264	+
	4265	+static const struct ib_device_ops mlx5_ib_dev_common_roce_ops = {
	4266	+ .create_rwq_ind_table = mlx5_ib_create_rwq_ind_table,
	4267	+ .create_wq = mlx5_ib_create_wq,
	4268	+ .destroy_rwq_ind_table = mlx5_ib_destroy_rwq_ind_table,
	4269	+ .destroy_wq = mlx5_ib_destroy_wq,
	4270	+ .get_netdev = mlx5_ib_get_netdev,
	4271	+ .modify_wq = mlx5_ib_modify_wq,
	4272	+
	4273	+ INIT_RDMA_OBJ_SIZE(ib_rwq_ind_table, mlx5_ib_rwq_ind_table,
	4274	+ ib_rwq_ind_tbl),
	4275	+};
	4276	+
	4277	+static int mlx5_ib_roce_init(struct mlx5_ib_dev *dev)
	4278	+{
	4279	+ struct mlx5_core_dev *mdev = dev->mdev;
	4280	+ enum rdma_link_layer ll;
	4281	+ int port_type_cap;
	4282	+ u8 port_num = 0;
	4283	+ int err;
	4284	+
	4285	+ port_type_cap = MLX5_CAP_GEN(mdev, port_type);
	4286	+ ll = mlx5_port_type_cap_to_rdma_ll(port_type_cap);
	4287	+
	4288	+ if (ll == IB_LINK_LAYER_ETHERNET) {
	4289	+ ib_set_device_ops(&dev->ib_dev, &mlx5_ib_dev_common_roce_ops);
	4290	+
	4291	+ port_num = mlx5_core_native_port_num(dev->mdev) - 1;
	4292	+
	4293	+ /* Register only for native ports */
	4294	+ err = mlx5_add_netdev_notifier(dev, port_num);
	4295	+ if (err \|\| dev->is_rep \|\| !mlx5_is_roce_enabled(mdev))
	4296	+ /*
	4297	+ * We don't enable ETH interface for
	4298	+ * 1. IB representors
	4299	+ * 2. User disabled ROCE through devlink interface
	4300	+ */
	4301	+ return err;
	4302	+
	4303	+ err = mlx5_enable_eth(dev);
	4304	+ if (err)
	4305	+ goto cleanup;
	4306	+ }
	4307	+
	4308	+ return 0;
	4309	+cleanup:
	4310	+ mlx5_remove_netdev_notifier(dev, port_num);
	4311	+ return err;
	4312	+}
	4313	+
	4314	+static void mlx5_ib_roce_cleanup(struct mlx5_ib_dev *dev)
	4315	+{
	4316	+ struct mlx5_core_dev *mdev = dev->mdev;
	4317	+ enum rdma_link_layer ll;
	4318	+ int port_type_cap;
	4319	+ u8 port_num;
	4320	+
	4321	+ port_type_cap = MLX5_CAP_GEN(mdev, port_type);
	4322	+ ll = mlx5_port_type_cap_to_rdma_ll(port_type_cap);
	4323	+
	4324	+ if (ll == IB_LINK_LAYER_ETHERNET) {
	4325	+ if (!dev->is_rep)
	4326	+ mlx5_disable_eth(dev);
	4327	+
	4328	+ port_num = mlx5_core_native_port_num(dev->mdev) - 1;
	4329	+ mlx5_remove_netdev_notifier(dev, port_num);
	4330	+ }
	4331	+}
	4332	+
	4333	+static int mlx5_ib_stage_cong_debugfs_init(struct mlx5_ib_dev *dev)
	4334	+{
	4335	+ mlx5_ib_init_cong_debugfs(dev,
	4336	+ mlx5_core_native_port_num(dev->mdev) - 1);
	4337	+ return 0;
	4338	+}
	4339	+
	4340	+static void mlx5_ib_stage_cong_debugfs_cleanup(struct mlx5_ib_dev *dev)
	4341	+{
	4342	+ mlx5_ib_cleanup_cong_debugfs(dev,
	4343	+ mlx5_core_native_port_num(dev->mdev) - 1);
	4344	+}
	4345	+
	4346	+static int mlx5_ib_stage_uar_init(struct mlx5_ib_dev *dev)
	4347	+{
	4348	+ dev->mdev->priv.uar = mlx5_get_uars_page(dev->mdev);
	4349	+ return PTR_ERR_OR_ZERO(dev->mdev->priv.uar);
	4350	+}
	4351	+
	4352	+static void mlx5_ib_stage_uar_cleanup(struct mlx5_ib_dev *dev)
	4353	+{
	4354	+ mlx5_put_uars_page(dev->mdev, dev->mdev->priv.uar);
	4355	+}
	4356	+
	4357	+static int mlx5_ib_stage_bfrag_init(struct mlx5_ib_dev *dev)
	4358	+{
	4359	+ int err;
	4360	+
	4361	+ err = mlx5_alloc_bfreg(dev->mdev, &dev->bfreg, false, false);
	4362	+ if (err)
	4363	+ return err;
	4364	+
	4365	+ err = mlx5_alloc_bfreg(dev->mdev, &dev->fp_bfreg, false, true);
	4366	+ if (err)
	4367	+ mlx5_free_bfreg(dev->mdev, &dev->bfreg);
	4368	+
	4369	+ return err;
	4370	+}
	4371	+
	4372	+static void mlx5_ib_stage_bfrag_cleanup(struct mlx5_ib_dev *dev)
	4373	+{
	4374	+ mlx5_free_bfreg(dev->mdev, &dev->fp_bfreg);
	4375	+ mlx5_free_bfreg(dev->mdev, &dev->bfreg);
	4376	+}
	4377	+
	4378	+static int mlx5_ib_stage_ib_reg_init(struct mlx5_ib_dev *dev)
	4379	+{
	4380	+ const char *name;
	4381	+
	4382	+ rdma_set_device_sysfs_group(&dev->ib_dev, &mlx5_attr_group);
	4383	+ if (!mlx5_lag_is_roce(dev->mdev))
	4384	+ name = "mlx5_%d";
	4385	+ else
	4386	+ name = "mlx5_bond_%d";
	4387	+ return ib_register_device(&dev->ib_dev, name, &dev->mdev->pdev->dev);
	4388	+}
	4389	+
	4390	+static void mlx5_ib_stage_pre_ib_reg_umr_cleanup(struct mlx5_ib_dev *dev)
4405	4391	{
4406	4392	int err;
4407	4393
..	..	@@ -4410,18 +4396,23 @@
4410	4396	mlx5_ib_warn(dev, "mr cache cleanup failed\n");
4411	4397
4412	4398	if (dev->umrc.qp)
4413		- mlx5_ib_destroy_qp(dev->umrc.qp);
	4399	+ mlx5_ib_destroy_qp(dev->umrc.qp, NULL);
4414	4400	if (dev->umrc.cq)
4415	4401	ib_free_cq(dev->umrc.cq);
4416	4402	if (dev->umrc.pd)
4417	4403	ib_dealloc_pd(dev->umrc.pd);
4418	4404	}
4419	4405
	4406	+static void mlx5_ib_stage_ib_reg_cleanup(struct mlx5_ib_dev *dev)
	4407	+{
	4408	+ ib_unregister_device(&dev->ib_dev);
	4409	+}
	4410	+
4420	4411	enum {
4421	4412	MAX_UMR_WR = 128,
4422	4413	};
4423	4414
4424		-static int create_umr_res(struct mlx5_ib_dev *dev)
	4415	+static int mlx5_ib_stage_post_ib_reg_umr_init(struct mlx5_ib_dev *dev)
4425	4416	{
4426	4417	struct ib_qp_init_attr *init_attr = NULL;
4427	4418	struct ib_qp_attr *attr = NULL;
..	..	@@ -4515,7 +4506,7 @@
4515	4506	return 0;
4516	4507
4517	4508	error_4:
4518		- mlx5_ib_destroy_qp(qp);
	4509	+ mlx5_ib_destroy_qp(qp, NULL);
4519	4510	dev->umrc.qp = NULL;
4520	4511
4521	4512	error_3:
..	..	@@ -4532,789 +4523,12 @@
4532	4523	return ret;
4533	4524	}
4534	4525
4535		-static u8 mlx5_get_umr_fence(u8 umr_fence_cap)
	4526	+static int mlx5_ib_stage_delay_drop_init(struct mlx5_ib_dev *dev)
4536	4527	{
4537		- switch (umr_fence_cap) {
4538		- case MLX5_CAP_UMR_FENCE_NONE:
4539		- return MLX5_FENCE_MODE_NONE;
4540		- case MLX5_CAP_UMR_FENCE_SMALL:
4541		- return MLX5_FENCE_MODE_INITIATOR_SMALL;
4542		- default:
4543		- return MLX5_FENCE_MODE_STRONG_ORDERING;
4544		- }
4545		-}
4546		-
4547		-static int create_dev_resources(struct mlx5_ib_resources *devr)
4548		-{
4549		- struct ib_srq_init_attr attr;
4550		- struct mlx5_ib_dev *dev;
4551		- struct ib_cq_init_attr cq_attr = {.cqe = 1};
4552		- int port;
4553		- int ret = 0;
4554		-
4555		- dev = container_of(devr, struct mlx5_ib_dev, devr);
4556		-
4557		- mutex_init(&devr->mutex);
4558		-
4559		- devr->p0 = mlx5_ib_alloc_pd(&dev->ib_dev, NULL, NULL);
4560		- if (IS_ERR(devr->p0)) {
4561		- ret = PTR_ERR(devr->p0);
4562		- goto error0;
4563		- }
4564		- devr->p0->device = &dev->ib_dev;
4565		- devr->p0->uobject = NULL;
4566		- atomic_set(&devr->p0->usecnt, 0);
4567		-
4568		- devr->c0 = mlx5_ib_create_cq(&dev->ib_dev, &cq_attr, NULL, NULL);
4569		- if (IS_ERR(devr->c0)) {
4570		- ret = PTR_ERR(devr->c0);
4571		- goto error1;
4572		- }
4573		- devr->c0->device = &dev->ib_dev;
4574		- devr->c0->uobject = NULL;
4575		- devr->c0->comp_handler = NULL;
4576		- devr->c0->event_handler = NULL;
4577		- devr->c0->cq_context = NULL;
4578		- atomic_set(&devr->c0->usecnt, 0);
4579		-
4580		- devr->x0 = mlx5_ib_alloc_xrcd(&dev->ib_dev, NULL, NULL);
4581		- if (IS_ERR(devr->x0)) {
4582		- ret = PTR_ERR(devr->x0);
4583		- goto error2;
4584		- }
4585		- devr->x0->device = &dev->ib_dev;
4586		- devr->x0->inode = NULL;
4587		- atomic_set(&devr->x0->usecnt, 0);
4588		- mutex_init(&devr->x0->tgt_qp_mutex);
4589		- INIT_LIST_HEAD(&devr->x0->tgt_qp_list);
4590		-
4591		- devr->x1 = mlx5_ib_alloc_xrcd(&dev->ib_dev, NULL, NULL);
4592		- if (IS_ERR(devr->x1)) {
4593		- ret = PTR_ERR(devr->x1);
4594		- goto error3;
4595		- }
4596		- devr->x1->device = &dev->ib_dev;
4597		- devr->x1->inode = NULL;
4598		- atomic_set(&devr->x1->usecnt, 0);
4599		- mutex_init(&devr->x1->tgt_qp_mutex);
4600		- INIT_LIST_HEAD(&devr->x1->tgt_qp_list);
4601		-
4602		- memset(&attr, 0, sizeof(attr));
4603		- attr.attr.max_sge = 1;
4604		- attr.attr.max_wr = 1;
4605		- attr.srq_type = IB_SRQT_XRC;
4606		- attr.ext.cq = devr->c0;
4607		- attr.ext.xrc.xrcd = devr->x0;
4608		-
4609		- devr->s0 = mlx5_ib_create_srq(devr->p0, &attr, NULL);
4610		- if (IS_ERR(devr->s0)) {
4611		- ret = PTR_ERR(devr->s0);
4612		- goto error4;
4613		- }
4614		- devr->s0->device = &dev->ib_dev;
4615		- devr->s0->pd = devr->p0;
4616		- devr->s0->uobject = NULL;
4617		- devr->s0->event_handler = NULL;
4618		- devr->s0->srq_context = NULL;
4619		- devr->s0->srq_type = IB_SRQT_XRC;
4620		- devr->s0->ext.xrc.xrcd = devr->x0;
4621		- devr->s0->ext.cq = devr->c0;
4622		- atomic_inc(&devr->s0->ext.xrc.xrcd->usecnt);
4623		- atomic_inc(&devr->s0->ext.cq->usecnt);
4624		- atomic_inc(&devr->p0->usecnt);
4625		- atomic_set(&devr->s0->usecnt, 0);
4626		-
4627		- memset(&attr, 0, sizeof(attr));
4628		- attr.attr.max_sge = 1;
4629		- attr.attr.max_wr = 1;
4630		- attr.srq_type = IB_SRQT_BASIC;
4631		- devr->s1 = mlx5_ib_create_srq(devr->p0, &attr, NULL);
4632		- if (IS_ERR(devr->s1)) {
4633		- ret = PTR_ERR(devr->s1);
4634		- goto error5;
4635		- }
4636		- devr->s1->device = &dev->ib_dev;
4637		- devr->s1->pd = devr->p0;
4638		- devr->s1->uobject = NULL;
4639		- devr->s1->event_handler = NULL;
4640		- devr->s1->srq_context = NULL;
4641		- devr->s1->srq_type = IB_SRQT_BASIC;
4642		- devr->s1->ext.cq = devr->c0;
4643		- atomic_inc(&devr->p0->usecnt);
4644		- atomic_set(&devr->s1->usecnt, 0);
4645		-
4646		- for (port = 0; port < ARRAY_SIZE(devr->ports); ++port) {
4647		- INIT_WORK(&devr->ports[port].pkey_change_work,
4648		- pkey_change_handler);
4649		- devr->ports[port].devr = devr;
4650		- }
4651		-
4652		- return 0;
4653		-
4654		-error5:
4655		- mlx5_ib_destroy_srq(devr->s0);
4656		-error4:
4657		- mlx5_ib_dealloc_xrcd(devr->x1);
4658		-error3:
4659		- mlx5_ib_dealloc_xrcd(devr->x0);
4660		-error2:
4661		- mlx5_ib_destroy_cq(devr->c0);
4662		-error1:
4663		- mlx5_ib_dealloc_pd(devr->p0);
4664		-error0:
4665		- return ret;
4666		-}
4667		-
4668		-static void destroy_dev_resources(struct mlx5_ib_resources *devr)
4669		-{
4670		- struct mlx5_ib_dev *dev =
4671		- container_of(devr, struct mlx5_ib_dev, devr);
4672		- int port;
4673		-
4674		- mlx5_ib_destroy_srq(devr->s1);
4675		- mlx5_ib_destroy_srq(devr->s0);
4676		- mlx5_ib_dealloc_xrcd(devr->x0);
4677		- mlx5_ib_dealloc_xrcd(devr->x1);
4678		- mlx5_ib_destroy_cq(devr->c0);
4679		- mlx5_ib_dealloc_pd(devr->p0);
4680		-
4681		- /* Make sure no change P_Key work items are still executing */
4682		- for (port = 0; port < dev->num_ports; ++port)
4683		- cancel_work_sync(&devr->ports[port].pkey_change_work);
4684		-}
4685		-
4686		-static u32 get_core_cap_flags(struct ib_device *ibdev,
4687		- struct mlx5_hca_vport_context *rep)
4688		-{
4689		- struct mlx5_ib_dev *dev = to_mdev(ibdev);
4690		- enum rdma_link_layer ll = mlx5_ib_port_link_layer(ibdev, 1);
4691		- u8 l3_type_cap = MLX5_CAP_ROCE(dev->mdev, l3_type);
4692		- u8 roce_version_cap = MLX5_CAP_ROCE(dev->mdev, roce_version);
4693		- bool raw_support = !mlx5_core_mp_enabled(dev->mdev);
4694		- u32 ret = 0;
4695		-
4696		- if (rep->grh_required)
4697		- ret \|= RDMA_CORE_CAP_IB_GRH_REQUIRED;
4698		-
4699		- if (ll == IB_LINK_LAYER_INFINIBAND)
4700		- return ret \| RDMA_CORE_PORT_IBA_IB;
4701		-
4702		- if (raw_support)
4703		- ret \|= RDMA_CORE_PORT_RAW_PACKET;
4704		-
4705		- if (!(l3_type_cap & MLX5_ROCE_L3_TYPE_IPV4_CAP))
4706		- return ret;
4707		-
4708		- if (!(l3_type_cap & MLX5_ROCE_L3_TYPE_IPV6_CAP))
4709		- return ret;
4710		-
4711		- if (roce_version_cap & MLX5_ROCE_VERSION_1_CAP)
4712		- ret \|= RDMA_CORE_PORT_IBA_ROCE;
4713		-
4714		- if (roce_version_cap & MLX5_ROCE_VERSION_2_CAP)
4715		- ret \|= RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
4716		-
4717		- return ret;
4718		-}
4719		-
4720		-static int mlx5_port_immutable(struct ib_device *ibdev, u8 port_num,
4721		- struct ib_port_immutable *immutable)
4722		-{
4723		- struct ib_port_attr attr;
4724		- struct mlx5_ib_dev *dev = to_mdev(ibdev);
4725		- enum rdma_link_layer ll = mlx5_ib_port_link_layer(ibdev, port_num);
4726		- struct mlx5_hca_vport_context rep = {0};
4727		- int err;
4728		-
4729		- err = ib_query_port(ibdev, port_num, &attr);
4730		- if (err)
4731		- return err;
4732		-
4733		- if (ll == IB_LINK_LAYER_INFINIBAND) {
4734		- err = mlx5_query_hca_vport_context(dev->mdev, 0, port_num, 0,
4735		- &rep);
4736		- if (err)
4737		- return err;
4738		- }
4739		-
4740		- immutable->pkey_tbl_len = attr.pkey_tbl_len;
4741		- immutable->gid_tbl_len = attr.gid_tbl_len;
4742		- immutable->core_cap_flags = get_core_cap_flags(ibdev, &rep);
4743		- if ((ll == IB_LINK_LAYER_INFINIBAND) \|\| MLX5_CAP_GEN(dev->mdev, roce))
4744		- immutable->max_mad_size = IB_MGMT_MAD_SIZE;
4745		-
4746		- return 0;
4747		-}
4748		-
4749		-static int mlx5_port_rep_immutable(struct ib_device *ibdev, u8 port_num,
4750		- struct ib_port_immutable *immutable)
4751		-{
4752		- struct ib_port_attr attr;
4753		- int err;
4754		-
4755		- immutable->core_cap_flags = RDMA_CORE_PORT_RAW_PACKET;
4756		-
4757		- err = ib_query_port(ibdev, port_num, &attr);
4758		- if (err)
4759		- return err;
4760		-
4761		- immutable->pkey_tbl_len = attr.pkey_tbl_len;
4762		- immutable->gid_tbl_len = attr.gid_tbl_len;
4763		- immutable->core_cap_flags = RDMA_CORE_PORT_RAW_PACKET;
4764		-
4765		- return 0;
4766		-}
4767		-
4768		-static void get_dev_fw_str(struct ib_device ibdev, char str)
4769		-{
4770		- struct mlx5_ib_dev *dev =
4771		- container_of(ibdev, struct mlx5_ib_dev, ib_dev);
4772		- snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%04d",
4773		- fw_rev_maj(dev->mdev), fw_rev_min(dev->mdev),
4774		- fw_rev_sub(dev->mdev));
4775		-}
4776		-
4777		-static int mlx5_eth_lag_init(struct mlx5_ib_dev *dev)
4778		-{
4779		- struct mlx5_core_dev *mdev = dev->mdev;
4780		- struct mlx5_flow_namespace *ns = mlx5_get_flow_namespace(mdev,
4781		- MLX5_FLOW_NAMESPACE_LAG);
4782		- struct mlx5_flow_table *ft;
4783		- int err;
4784		-
4785		- if (!ns \|\| !mlx5_lag_is_active(mdev))
4786		- return 0;
4787		-
4788		- err = mlx5_cmd_create_vport_lag(mdev);
4789		- if (err)
4790		- return err;
4791		-
4792		- ft = mlx5_create_lag_demux_flow_table(ns, 0, 0);
4793		- if (IS_ERR(ft)) {
4794		- err = PTR_ERR(ft);
4795		- goto err_destroy_vport_lag;
4796		- }
4797		-
4798		- dev->flow_db->lag_demux_ft = ft;
4799		- return 0;
4800		-
4801		-err_destroy_vport_lag:
4802		- mlx5_cmd_destroy_vport_lag(mdev);
4803		- return err;
4804		-}
4805		-
4806		-static void mlx5_eth_lag_cleanup(struct mlx5_ib_dev *dev)
4807		-{
4808		- struct mlx5_core_dev *mdev = dev->mdev;
4809		-
4810		- if (dev->flow_db->lag_demux_ft) {
4811		- mlx5_destroy_flow_table(dev->flow_db->lag_demux_ft);
4812		- dev->flow_db->lag_demux_ft = NULL;
4813		-
4814		- mlx5_cmd_destroy_vport_lag(mdev);
4815		- }
4816		-}
4817		-
4818		-static int mlx5_add_netdev_notifier(struct mlx5_ib_dev *dev, u8 port_num)
4819		-{
4820		- int err;
4821		-
4822		- dev->roce[port_num].nb.notifier_call = mlx5_netdev_event;
4823		- err = register_netdevice_notifier(&dev->roce[port_num].nb);
4824		- if (err) {
4825		- dev->roce[port_num].nb.notifier_call = NULL;
4826		- return err;
4827		- }
4828		-
4829		- return 0;
4830		-}
4831		-
4832		-static void mlx5_remove_netdev_notifier(struct mlx5_ib_dev *dev, u8 port_num)
4833		-{
4834		- if (dev->roce[port_num].nb.notifier_call) {
4835		- unregister_netdevice_notifier(&dev->roce[port_num].nb);
4836		- dev->roce[port_num].nb.notifier_call = NULL;
4837		- }
4838		-}
4839		-
4840		-static int mlx5_enable_eth(struct mlx5_ib_dev *dev)
4841		-{
4842		- int err;
4843		-
4844		- if (MLX5_CAP_GEN(dev->mdev, roce)) {
4845		- err = mlx5_nic_vport_enable_roce(dev->mdev);
4846		- if (err)
4847		- return err;
4848		- }
4849		-
4850		- err = mlx5_eth_lag_init(dev);
4851		- if (err)
4852		- goto err_disable_roce;
4853		-
4854		- return 0;
4855		-
4856		-err_disable_roce:
4857		- if (MLX5_CAP_GEN(dev->mdev, roce))
4858		- mlx5_nic_vport_disable_roce(dev->mdev);
4859		-
4860		- return err;
4861		-}
4862		-
4863		-static void mlx5_disable_eth(struct mlx5_ib_dev *dev)
4864		-{
4865		- mlx5_eth_lag_cleanup(dev);
4866		- if (MLX5_CAP_GEN(dev->mdev, roce))
4867		- mlx5_nic_vport_disable_roce(dev->mdev);
4868		-}
4869		-
4870		-struct mlx5_ib_counter {
4871		- const char *name;
4872		- size_t offset;
4873		-};
4874		-
4875		-#define INIT_Q_COUNTER(_name) \
4876		- { .name = #_name, .offset = MLX5_BYTE_OFF(query_q_counter_out, _name)}
4877		-
4878		-static const struct mlx5_ib_counter basic_q_cnts[] = {
4879		- INIT_Q_COUNTER(rx_write_requests),
4880		- INIT_Q_COUNTER(rx_read_requests),
4881		- INIT_Q_COUNTER(rx_atomic_requests),
4882		- INIT_Q_COUNTER(out_of_buffer),
4883		-};
4884		-
4885		-static const struct mlx5_ib_counter out_of_seq_q_cnts[] = {
4886		- INIT_Q_COUNTER(out_of_sequence),
4887		-};
4888		-
4889		-static const struct mlx5_ib_counter retrans_q_cnts[] = {
4890		- INIT_Q_COUNTER(duplicate_request),
4891		- INIT_Q_COUNTER(rnr_nak_retry_err),
4892		- INIT_Q_COUNTER(packet_seq_err),
4893		- INIT_Q_COUNTER(implied_nak_seq_err),
4894		- INIT_Q_COUNTER(local_ack_timeout_err),
4895		-};
4896		-
4897		-#define INIT_CONG_COUNTER(_name) \
4898		- { .name = #_name, .offset = \
4899		- MLX5_BYTE_OFF(query_cong_statistics_out, _name ## _high)}
4900		-
4901		-static const struct mlx5_ib_counter cong_cnts[] = {
4902		- INIT_CONG_COUNTER(rp_cnp_ignored),
4903		- INIT_CONG_COUNTER(rp_cnp_handled),
4904		- INIT_CONG_COUNTER(np_ecn_marked_roce_packets),
4905		- INIT_CONG_COUNTER(np_cnp_sent),
4906		-};
4907		-
4908		-static const struct mlx5_ib_counter extended_err_cnts[] = {
4909		- INIT_Q_COUNTER(resp_local_length_error),
4910		- INIT_Q_COUNTER(resp_cqe_error),
4911		- INIT_Q_COUNTER(req_cqe_error),
4912		- INIT_Q_COUNTER(req_remote_invalid_request),
4913		- INIT_Q_COUNTER(req_remote_access_errors),
4914		- INIT_Q_COUNTER(resp_remote_access_errors),
4915		- INIT_Q_COUNTER(resp_cqe_flush_error),
4916		- INIT_Q_COUNTER(req_cqe_flush_error),
4917		-};
4918		-
4919		-#define INIT_EXT_PPCNT_COUNTER(_name) \
4920		- { .name = #_name, .offset = \
4921		- MLX5_BYTE_OFF(ppcnt_reg, \
4922		- counter_set.eth_extended_cntrs_grp_data_layout._name##_high)}
4923		-
4924		-static const struct mlx5_ib_counter ext_ppcnt_cnts[] = {
4925		- INIT_EXT_PPCNT_COUNTER(rx_icrc_encapsulated),
4926		-};
4927		-
4928		-static void mlx5_ib_dealloc_counters(struct mlx5_ib_dev *dev)
4929		-{
4930		- int i;
4931		-
4932		- for (i = 0; i < dev->num_ports; i++) {
4933		- if (dev->port[i].cnts.set_id_valid)
4934		- mlx5_core_dealloc_q_counter(dev->mdev,
4935		- dev->port[i].cnts.set_id);
4936		- kfree(dev->port[i].cnts.names);
4937		- kfree(dev->port[i].cnts.offsets);
4938		- }
4939		-}
4940		-
4941		-static int __mlx5_ib_alloc_counters(struct mlx5_ib_dev *dev,
4942		- struct mlx5_ib_counters *cnts)
4943		-{
4944		- u32 num_counters;
4945		-
4946		- num_counters = ARRAY_SIZE(basic_q_cnts);
4947		-
4948		- if (MLX5_CAP_GEN(dev->mdev, out_of_seq_cnt))
4949		- num_counters += ARRAY_SIZE(out_of_seq_q_cnts);
4950		-
4951		- if (MLX5_CAP_GEN(dev->mdev, retransmission_q_counters))
4952		- num_counters += ARRAY_SIZE(retrans_q_cnts);
4953		-
4954		- if (MLX5_CAP_GEN(dev->mdev, enhanced_error_q_counters))
4955		- num_counters += ARRAY_SIZE(extended_err_cnts);
4956		-
4957		- cnts->num_q_counters = num_counters;
4958		-
4959		- if (MLX5_CAP_GEN(dev->mdev, cc_query_allowed)) {
4960		- cnts->num_cong_counters = ARRAY_SIZE(cong_cnts);
4961		- num_counters += ARRAY_SIZE(cong_cnts);
4962		- }
4963		- if (MLX5_CAP_PCAM_FEATURE(dev->mdev, rx_icrc_encapsulated_counter)) {
4964		- cnts->num_ext_ppcnt_counters = ARRAY_SIZE(ext_ppcnt_cnts);
4965		- num_counters += ARRAY_SIZE(ext_ppcnt_cnts);
4966		- }
4967		- cnts->names = kcalloc(num_counters, sizeof(cnts->names), GFP_KERNEL);
4968		- if (!cnts->names)
4969		- return -ENOMEM;
4970		-
4971		- cnts->offsets = kcalloc(num_counters,
4972		- sizeof(cnts->offsets), GFP_KERNEL);
4973		- if (!cnts->offsets)
4974		- goto err_names;
4975		-
4976		- return 0;
4977		-
4978		-err_names:
4979		- kfree(cnts->names);
4980		- cnts->names = NULL;
4981		- return -ENOMEM;
4982		-}
4983		-
4984		-static void mlx5_ib_fill_counters(struct mlx5_ib_dev *dev,
4985		- const char **names,
4986		- size_t *offsets)
4987		-{
4988		- int i;
4989		- int j = 0;
4990		-
4991		- for (i = 0; i < ARRAY_SIZE(basic_q_cnts); i++, j++) {
4992		- names[j] = basic_q_cnts[i].name;
4993		- offsets[j] = basic_q_cnts[i].offset;
4994		- }
4995		-
4996		- if (MLX5_CAP_GEN(dev->mdev, out_of_seq_cnt)) {
4997		- for (i = 0; i < ARRAY_SIZE(out_of_seq_q_cnts); i++, j++) {
4998		- names[j] = out_of_seq_q_cnts[i].name;
4999		- offsets[j] = out_of_seq_q_cnts[i].offset;
5000		- }
5001		- }
5002		-
5003		- if (MLX5_CAP_GEN(dev->mdev, retransmission_q_counters)) {
5004		- for (i = 0; i < ARRAY_SIZE(retrans_q_cnts); i++, j++) {
5005		- names[j] = retrans_q_cnts[i].name;
5006		- offsets[j] = retrans_q_cnts[i].offset;
5007		- }
5008		- }
5009		-
5010		- if (MLX5_CAP_GEN(dev->mdev, enhanced_error_q_counters)) {
5011		- for (i = 0; i < ARRAY_SIZE(extended_err_cnts); i++, j++) {
5012		- names[j] = extended_err_cnts[i].name;
5013		- offsets[j] = extended_err_cnts[i].offset;
5014		- }
5015		- }
5016		-
5017		- if (MLX5_CAP_GEN(dev->mdev, cc_query_allowed)) {
5018		- for (i = 0; i < ARRAY_SIZE(cong_cnts); i++, j++) {
5019		- names[j] = cong_cnts[i].name;
5020		- offsets[j] = cong_cnts[i].offset;
5021		- }
5022		- }
5023		-
5024		- if (MLX5_CAP_PCAM_FEATURE(dev->mdev, rx_icrc_encapsulated_counter)) {
5025		- for (i = 0; i < ARRAY_SIZE(ext_ppcnt_cnts); i++, j++) {
5026		- names[j] = ext_ppcnt_cnts[i].name;
5027		- offsets[j] = ext_ppcnt_cnts[i].offset;
5028		- }
5029		- }
5030		-}
5031		-
5032		-static int mlx5_ib_alloc_counters(struct mlx5_ib_dev *dev)
5033		-{
5034		- int err = 0;
5035		- int i;
5036		-
5037		- for (i = 0; i < dev->num_ports; i++) {
5038		- err = __mlx5_ib_alloc_counters(dev, &dev->port[i].cnts);
5039		- if (err)
5040		- goto err_alloc;
5041		-
5042		- mlx5_ib_fill_counters(dev, dev->port[i].cnts.names,
5043		- dev->port[i].cnts.offsets);
5044		-
5045		- err = mlx5_core_alloc_q_counter(dev->mdev,
5046		- &dev->port[i].cnts.set_id);
5047		- if (err) {
5048		- mlx5_ib_warn(dev,
5049		- "couldn't allocate queue counter for port %d, err %d\n",
5050		- i + 1, err);
5051		- goto err_alloc;
5052		- }
5053		- dev->port[i].cnts.set_id_valid = true;
5054		- }
5055		-
5056		- return 0;
5057		-
5058		-err_alloc:
5059		- mlx5_ib_dealloc_counters(dev);
5060		- return err;
5061		-}
5062		-
5063		-static struct rdma_hw_stats mlx5_ib_alloc_hw_stats(struct ib_device ibdev,
5064		- u8 port_num)
5065		-{
5066		- struct mlx5_ib_dev *dev = to_mdev(ibdev);
5067		- struct mlx5_ib_port *port = &dev->port[port_num - 1];
5068		-
5069		- /* We support only per port stats */
5070		- if (port_num == 0)
5071		- return NULL;
5072		-
5073		- return rdma_alloc_hw_stats_struct(port->cnts.names,
5074		- port->cnts.num_q_counters +
5075		- port->cnts.num_cong_counters +
5076		- port->cnts.num_ext_ppcnt_counters,
5077		- RDMA_HW_STATS_DEFAULT_LIFESPAN);
5078		-}
5079		-
5080		-static int mlx5_ib_query_q_counters(struct mlx5_core_dev *mdev,
5081		- struct mlx5_ib_port *port,
5082		- struct rdma_hw_stats *stats)
5083		-{
5084		- int outlen = MLX5_ST_SZ_BYTES(query_q_counter_out);
5085		- void *out;
5086		- __be32 val;
5087		- int ret, i;
5088		-
5089		- out = kvzalloc(outlen, GFP_KERNEL);
5090		- if (!out)
5091		- return -ENOMEM;
5092		-
5093		- ret = mlx5_core_query_q_counter(mdev,
5094		- port->cnts.set_id, 0,
5095		- out, outlen);
5096		- if (ret)
5097		- goto free;
5098		-
5099		- for (i = 0; i < port->cnts.num_q_counters; i++) {
5100		- val = (__be32 )(out + port->cnts.offsets[i]);
5101		- stats->value[i] = (u64)be32_to_cpu(val);
5102		- }
5103		-
5104		-free:
5105		- kvfree(out);
5106		- return ret;
5107		-}
5108		-
5109		-static int mlx5_ib_query_ext_ppcnt_counters(struct mlx5_ib_dev *dev,
5110		- struct mlx5_ib_port *port,
5111		- struct rdma_hw_stats *stats)
5112		-{
5113		- int offset = port->cnts.num_q_counters + port->cnts.num_cong_counters;
5114		- int sz = MLX5_ST_SZ_BYTES(ppcnt_reg);
5115		- int ret, i;
5116		- void *out;
5117		-
5118		- out = kvzalloc(sz, GFP_KERNEL);
5119		- if (!out)
5120		- return -ENOMEM;
5121		-
5122		- ret = mlx5_cmd_query_ext_ppcnt_counters(dev->mdev, out);
5123		- if (ret)
5124		- goto free;
5125		-
5126		- for (i = 0; i < port->cnts.num_ext_ppcnt_counters; i++) {
5127		- stats->value[i + offset] =
5128		- be64_to_cpup((__be64 *)(out +
5129		- port->cnts.offsets[i + offset]));
5130		- }
5131		-
5132		-free:
5133		- kvfree(out);
5134		- return ret;
5135		-}
5136		-
5137		-static int mlx5_ib_get_hw_stats(struct ib_device *ibdev,
5138		- struct rdma_hw_stats *stats,
5139		- u8 port_num, int index)
5140		-{
5141		- struct mlx5_ib_dev *dev = to_mdev(ibdev);
5142		- struct mlx5_ib_port *port = &dev->port[port_num - 1];
5143		- struct mlx5_core_dev *mdev;
5144		- int ret, num_counters;
5145		- u8 mdev_port_num;
5146		-
5147		- if (!stats)
5148		- return -EINVAL;
5149		-
5150		- num_counters = port->cnts.num_q_counters +
5151		- port->cnts.num_cong_counters +
5152		- port->cnts.num_ext_ppcnt_counters;
5153		-
5154		- /* q_counters are per IB device, query the master mdev */
5155		- ret = mlx5_ib_query_q_counters(dev->mdev, port, stats);
5156		- if (ret)
5157		- return ret;
5158		-
5159		- if (MLX5_CAP_PCAM_FEATURE(dev->mdev, rx_icrc_encapsulated_counter)) {
5160		- ret = mlx5_ib_query_ext_ppcnt_counters(dev, port, stats);
5161		- if (ret)
5162		- return ret;
5163		- }
5164		-
5165		- if (MLX5_CAP_GEN(dev->mdev, cc_query_allowed)) {
5166		- mdev = mlx5_ib_get_native_port_mdev(dev, port_num,
5167		- &mdev_port_num);
5168		- if (!mdev) {
5169		- /* If port is not affiliated yet, its in down state
5170		- * which doesn't have any counters yet, so it would be
5171		- * zero. So no need to read from the HCA.
5172		- */
5173		- goto done;
5174		- }
5175		- ret = mlx5_lag_query_cong_counters(dev->mdev,
5176		- stats->value +
5177		- port->cnts.num_q_counters,
5178		- port->cnts.num_cong_counters,
5179		- port->cnts.offsets +
5180		- port->cnts.num_q_counters);
5181		-
5182		- mlx5_ib_put_native_port_mdev(dev, port_num);
5183		- if (ret)
5184		- return ret;
5185		- }
5186		-
5187		-done:
5188		- return num_counters;
5189		-}
5190		-
5191		-static struct net_device*
5192		-mlx5_ib_alloc_rdma_netdev(struct ib_device *hca,
5193		- u8 port_num,
5194		- enum rdma_netdev_t type,
5195		- const char *name,
5196		- unsigned char name_assign_type,
5197		- void (setup)(struct net_device ))
5198		-{
5199		- struct net_device *netdev;
5200		-
5201		- if (type != RDMA_NETDEV_IPOIB)
5202		- return ERR_PTR(-EOPNOTSUPP);
5203		-
5204		- netdev = mlx5_rdma_netdev_alloc(to_mdev(hca)->mdev, hca,
5205		- name, setup);
5206		- return netdev;
5207		-}
5208		-
5209		-static void delay_drop_debugfs_cleanup(struct mlx5_ib_dev *dev)
5210		-{
5211		- if (!dev->delay_drop.dbg)
5212		- return;
5213		- debugfs_remove_recursive(dev->delay_drop.dbg->dir_debugfs);
5214		- kfree(dev->delay_drop.dbg);
5215		- dev->delay_drop.dbg = NULL;
5216		-}
	4528	+ struct dentry *root;
5217	4529
5218		-static void cancel_delay_drop(struct mlx5_ib_dev *dev)
5219		-{
5220	4530	if (!(dev->ib_dev.attrs.raw_packet_caps & IB_RAW_PACKET_CAP_DELAY_DROP))
5221		- return;
5222		-
5223		- cancel_work_sync(&dev->delay_drop.delay_drop_work);
5224		- delay_drop_debugfs_cleanup(dev);
5225		-}
5226		-
5227		-static ssize_t delay_drop_timeout_read(struct file filp, char __user buf,
5228		- size_t count, loff_t *pos)
5229		-{
5230		- struct mlx5_ib_delay_drop *delay_drop = filp->private_data;
5231		- char lbuf[20];
5232		- int len;
5233		-
5234		- len = snprintf(lbuf, sizeof(lbuf), "%u\n", delay_drop->timeout);
5235		- return simple_read_from_buffer(buf, count, pos, lbuf, len);
5236		-}
5237		-
5238		-static ssize_t delay_drop_timeout_write(struct file filp, const char __user buf,
5239		- size_t count, loff_t *pos)
5240		-{
5241		- struct mlx5_ib_delay_drop *delay_drop = filp->private_data;
5242		- u32 timeout;
5243		- u32 var;
5244		-
5245		- if (kstrtouint_from_user(buf, count, 0, &var))
5246		- return -EFAULT;
5247		-
5248		- timeout = min_t(u32, roundup(var, 100), MLX5_MAX_DELAY_DROP_TIMEOUT_MS *
5249		- 1000);
5250		- if (timeout != var)
5251		- mlx5_ib_dbg(delay_drop->dev, "Round delay drop timeout to %u usec\n",
5252		- timeout);
5253		-
5254		- delay_drop->timeout = timeout;
5255		-
5256		- return count;
5257		-}
5258		-
5259		-static const struct file_operations fops_delay_drop_timeout = {
5260		- .owner = THIS_MODULE,
5261		- .open = simple_open,
5262		- .write = delay_drop_timeout_write,
5263		- .read = delay_drop_timeout_read,
5264		-};
5265		-
5266		-static int delay_drop_debugfs_init(struct mlx5_ib_dev *dev)
5267		-{
5268		- struct mlx5_ib_dbg_delay_drop *dbg;
5269		-
5270		- if (!mlx5_debugfs_root)
5271	4531	return 0;
5272		-
5273		- dbg = kzalloc(sizeof(*dbg), GFP_KERNEL);
5274		- if (!dbg)
5275		- return -ENOMEM;
5276		-
5277		- dev->delay_drop.dbg = dbg;
5278		-
5279		- dbg->dir_debugfs =
5280		- debugfs_create_dir("delay_drop",
5281		- dev->mdev->priv.dbg_root);
5282		- if (!dbg->dir_debugfs)
5283		- goto out_debugfs;
5284		-
5285		- dbg->events_cnt_debugfs =
5286		- debugfs_create_atomic_t("num_timeout_events", 0400,
5287		- dbg->dir_debugfs,
5288		- &dev->delay_drop.events_cnt);
5289		- if (!dbg->events_cnt_debugfs)
5290		- goto out_debugfs;
5291		-
5292		- dbg->rqs_cnt_debugfs =
5293		- debugfs_create_atomic_t("num_rqs", 0400,
5294		- dbg->dir_debugfs,
5295		- &dev->delay_drop.rqs_cnt);
5296		- if (!dbg->rqs_cnt_debugfs)
5297		- goto out_debugfs;
5298		-
5299		- dbg->timeout_debugfs =
5300		- debugfs_create_file("timeout", 0600,
5301		- dbg->dir_debugfs,
5302		- &dev->delay_drop,
5303		- &fops_delay_drop_timeout);
5304		- if (!dbg->timeout_debugfs)
5305		- goto out_debugfs;
5306		-
5307		- return 0;
5308		-
5309		-out_debugfs:
5310		- delay_drop_debugfs_cleanup(dev);
5311		- return -ENOMEM;
5312		-}
5313		-
5314		-static void init_delay_drop(struct mlx5_ib_dev *dev)
5315		-{
5316		- if (!(dev->ib_dev.attrs.raw_packet_caps & IB_RAW_PACKET_CAP_DELAY_DROP))
5317		- return;
5318	4532
5319	4533	mutex_init(&dev->delay_drop.lock);
5320	4534	dev->delay_drop.dev = dev;
..	..	@@ -5324,855 +4538,52 @@
5324	4538	atomic_set(&dev->delay_drop.rqs_cnt, 0);
5325	4539	atomic_set(&dev->delay_drop.events_cnt, 0);
5326	4540
5327		- if (delay_drop_debugfs_init(dev))
5328		- mlx5_ib_warn(dev, "Failed to init delay drop debugfs\n");
5329		-}
5330		-
5331		-static const struct cpumask *
5332		-mlx5_ib_get_vector_affinity(struct ib_device *ibdev, int comp_vector)
5333		-{
5334		- struct mlx5_ib_dev *dev = to_mdev(ibdev);
5335		-
5336		- return mlx5_get_vector_affinity_hint(dev->mdev, comp_vector);
5337		-}
5338		-
5339		-/* The mlx5_ib_multiport_mutex should be held when calling this function */
5340		-static void mlx5_ib_unbind_slave_port(struct mlx5_ib_dev *ibdev,
5341		- struct mlx5_ib_multiport_info *mpi)
5342		-{
5343		- u8 port_num = mlx5_core_native_port_num(mpi->mdev) - 1;
5344		- struct mlx5_ib_port *port = &ibdev->port[port_num];
5345		- int comps;
5346		- int err;
5347		- int i;
5348		-
5349		- mlx5_ib_cleanup_cong_debugfs(ibdev, port_num);
5350		-
5351		- spin_lock(&port->mp.mpi_lock);
5352		- if (!mpi->ibdev) {
5353		- spin_unlock(&port->mp.mpi_lock);
5354		- return;
5355		- }
5356		- mpi->ibdev = NULL;
5357		-
5358		- spin_unlock(&port->mp.mpi_lock);
5359		- mlx5_remove_netdev_notifier(ibdev, port_num);
5360		- spin_lock(&port->mp.mpi_lock);
5361		-
5362		- comps = mpi->mdev_refcnt;
5363		- if (comps) {
5364		- mpi->unaffiliate = true;
5365		- init_completion(&mpi->unref_comp);
5366		- spin_unlock(&port->mp.mpi_lock);
5367		-
5368		- for (i = 0; i < comps; i++)
5369		- wait_for_completion(&mpi->unref_comp);
5370		-
5371		- spin_lock(&port->mp.mpi_lock);
5372		- mpi->unaffiliate = false;
5373		- }
5374		-
5375		- port->mp.mpi = NULL;
5376		-
5377		- spin_unlock(&port->mp.mpi_lock);
5378		-
5379		- err = mlx5_nic_vport_unaffiliate_multiport(mpi->mdev);
5380		-
5381		- mlx5_ib_dbg(ibdev, "unaffiliated port %d\n", port_num + 1);
5382		- /* Log an error, still needed to cleanup the pointers and add
5383		- * it back to the list.
5384		- */
5385		- if (err)
5386		- mlx5_ib_err(ibdev, "Failed to unaffiliate port %u\n",
5387		- port_num + 1);
5388		-
5389		- ibdev->roce[port_num].last_port_state = IB_PORT_DOWN;
5390		-}
5391		-
5392		-/* The mlx5_ib_multiport_mutex should be held when calling this function */
5393		-static bool mlx5_ib_bind_slave_port(struct mlx5_ib_dev *ibdev,
5394		- struct mlx5_ib_multiport_info *mpi)
5395		-{
5396		- u8 port_num = mlx5_core_native_port_num(mpi->mdev) - 1;
5397		- int err;
5398		-
5399		- spin_lock(&ibdev->port[port_num].mp.mpi_lock);
5400		- if (ibdev->port[port_num].mp.mpi) {
5401		- mlx5_ib_dbg(ibdev, "port %d already affiliated.\n",
5402		- port_num + 1);
5403		- spin_unlock(&ibdev->port[port_num].mp.mpi_lock);
5404		- return false;
5405		- }
5406		-
5407		- ibdev->port[port_num].mp.mpi = mpi;
5408		- mpi->ibdev = ibdev;
5409		- spin_unlock(&ibdev->port[port_num].mp.mpi_lock);
5410		-
5411		- err = mlx5_nic_vport_affiliate_multiport(ibdev->mdev, mpi->mdev);
5412		- if (err)
5413		- goto unbind;
5414		-
5415		- err = get_port_caps(ibdev, mlx5_core_native_port_num(mpi->mdev));
5416		- if (err)
5417		- goto unbind;
5418		-
5419		- err = mlx5_add_netdev_notifier(ibdev, port_num);
5420		- if (err) {
5421		- mlx5_ib_err(ibdev, "failed adding netdev notifier for port %u\n",
5422		- port_num + 1);
5423		- goto unbind;
5424		- }
5425		-
5426		- err = mlx5_ib_init_cong_debugfs(ibdev, port_num);
5427		- if (err)
5428		- goto unbind;
5429		-
5430		- return true;
5431		-
5432		-unbind:
5433		- mlx5_ib_unbind_slave_port(ibdev, mpi);
5434		- return false;
5435		-}
5436		-
5437		-static int mlx5_ib_init_multiport_master(struct mlx5_ib_dev *dev)
5438		-{
5439		- int port_num = mlx5_core_native_port_num(dev->mdev) - 1;
5440		- enum rdma_link_layer ll = mlx5_ib_port_link_layer(&dev->ib_dev,
5441		- port_num + 1);
5442		- struct mlx5_ib_multiport_info *mpi;
5443		- int err;
5444		- int i;
5445		-
5446		- if (!mlx5_core_is_mp_master(dev->mdev) \|\| ll != IB_LINK_LAYER_ETHERNET)
	4541	+ if (!mlx5_debugfs_root)
5447	4542	return 0;
5448	4543
5449		- err = mlx5_query_nic_vport_system_image_guid(dev->mdev,
5450		- &dev->sys_image_guid);
5451		- if (err)
5452		- return err;
	4544	+ root = debugfs_create_dir("delay_drop", dev->mdev->priv.dbg_root);
	4545	+ dev->delay_drop.dir_debugfs = root;
5453	4546
5454		- err = mlx5_nic_vport_enable_roce(dev->mdev);
5455		- if (err)
5456		- return err;
5457		-
5458		- mutex_lock(&mlx5_ib_multiport_mutex);
5459		- for (i = 0; i < dev->num_ports; i++) {
5460		- bool bound = false;
5461		-
5462		- /* build a stub multiport info struct for the native port. */
5463		- if (i == port_num) {
5464		- mpi = kzalloc(sizeof(*mpi), GFP_KERNEL);
5465		- if (!mpi) {
5466		- mutex_unlock(&mlx5_ib_multiport_mutex);
5467		- mlx5_nic_vport_disable_roce(dev->mdev);
5468		- return -ENOMEM;
5469		- }
5470		-
5471		- mpi->is_master = true;
5472		- mpi->mdev = dev->mdev;
5473		- mpi->sys_image_guid = dev->sys_image_guid;
5474		- dev->port[i].mp.mpi = mpi;
5475		- mpi->ibdev = dev;
5476		- mpi = NULL;
5477		- continue;
5478		- }
5479		-
5480		- list_for_each_entry(mpi, &mlx5_ib_unaffiliated_port_list,
5481		- list) {
5482		- if (dev->sys_image_guid == mpi->sys_image_guid &&
5483		- (mlx5_core_native_port_num(mpi->mdev) - 1) == i) {
5484		- bound = mlx5_ib_bind_slave_port(dev, mpi);
5485		- }
5486		-
5487		- if (bound) {
5488		- dev_dbg(&mpi->mdev->pdev->dev, "removing port from unaffiliated list.\n");
5489		- mlx5_ib_dbg(dev, "port %d bound\n", i + 1);
5490		- list_del(&mpi->list);
5491		- break;
5492		- }
5493		- }
5494		- if (!bound) {
5495		- get_port_caps(dev, i + 1);
5496		- mlx5_ib_dbg(dev, "no free port found for port %d\n",
5497		- i + 1);
5498		- }
5499		- }
5500		-
5501		- list_add_tail(&dev->ib_dev_list, &mlx5_ib_dev_list);
5502		- mutex_unlock(&mlx5_ib_multiport_mutex);
5503		- return err;
5504		-}
5505		-
5506		-static void mlx5_ib_cleanup_multiport_master(struct mlx5_ib_dev *dev)
5507		-{
5508		- int port_num = mlx5_core_native_port_num(dev->mdev) - 1;
5509		- enum rdma_link_layer ll = mlx5_ib_port_link_layer(&dev->ib_dev,
5510		- port_num + 1);
5511		- int i;
5512		-
5513		- if (!mlx5_core_is_mp_master(dev->mdev) \|\| ll != IB_LINK_LAYER_ETHERNET)
5514		- return;
5515		-
5516		- mutex_lock(&mlx5_ib_multiport_mutex);
5517		- for (i = 0; i < dev->num_ports; i++) {
5518		- if (dev->port[i].mp.mpi) {
5519		- /* Destroy the native port stub */
5520		- if (i == port_num) {
5521		- kfree(dev->port[i].mp.mpi);
5522		- dev->port[i].mp.mpi = NULL;
5523		- } else {
5524		- mlx5_ib_dbg(dev, "unbinding port_num: %d\n", i + 1);
5525		- list_add_tail(&dev->port[i].mp.mpi->list,
5526		- &mlx5_ib_unaffiliated_port_list);
5527		- mlx5_ib_unbind_slave_port(dev, dev->port[i].mp.mpi);
5528		- }
5529		- }
5530		- }
5531		-
5532		- mlx5_ib_dbg(dev, "removing from devlist\n");
5533		- list_del(&dev->ib_dev_list);
5534		- mutex_unlock(&mlx5_ib_multiport_mutex);
5535		-
5536		- mlx5_nic_vport_disable_roce(dev->mdev);
5537		-}
5538		-
5539		-ADD_UVERBS_ATTRIBUTES_SIMPLE(
5540		- mlx5_ib_dm,
5541		- UVERBS_OBJECT_DM,
5542		- UVERBS_METHOD_DM_ALLOC,
5543		- UVERBS_ATTR_PTR_OUT(MLX5_IB_ATTR_ALLOC_DM_RESP_START_OFFSET,
5544		- UVERBS_ATTR_TYPE(u64),
5545		- UA_MANDATORY),
5546		- UVERBS_ATTR_PTR_OUT(MLX5_IB_ATTR_ALLOC_DM_RESP_PAGE_INDEX,
5547		- UVERBS_ATTR_TYPE(u16),
5548		- UA_MANDATORY));
5549		-
5550		-ADD_UVERBS_ATTRIBUTES_SIMPLE(
5551		- mlx5_ib_flow_action,
5552		- UVERBS_OBJECT_FLOW_ACTION,
5553		- UVERBS_METHOD_FLOW_ACTION_ESP_CREATE,
5554		- UVERBS_ATTR_FLAGS_IN(MLX5_IB_ATTR_CREATE_FLOW_ACTION_FLAGS,
5555		- enum mlx5_ib_uapi_flow_action_flags));
5556		-
5557		-static int populate_specs_root(struct mlx5_ib_dev *dev)
5558		-{
5559		- const struct uverbs_object_tree_def **trees = dev->driver_trees;
5560		- size_t num_trees = 0;
5561		-
5562		- if (mlx5_accel_ipsec_device_caps(dev->mdev) &
5563		- MLX5_ACCEL_IPSEC_CAP_DEVICE)
5564		- trees[num_trees++] = &mlx5_ib_flow_action;
5565		-
5566		- if (MLX5_CAP_DEV_MEM(dev->mdev, memic))
5567		- trees[num_trees++] = &mlx5_ib_dm;
5568		-
5569		- if (MLX5_CAP_GEN_64(dev->mdev, general_obj_types) &
5570		- MLX5_GENERAL_OBJ_TYPES_CAP_UCTX)
5571		- trees[num_trees++] = mlx5_ib_get_devx_tree();
5572		-
5573		- num_trees += mlx5_ib_get_flow_trees(trees + num_trees);
5574		-
5575		- WARN_ON(num_trees >= ARRAY_SIZE(dev->driver_trees));
5576		- trees[num_trees] = NULL;
5577		- dev->ib_dev.driver_specs = trees;
5578		-
5579		- return 0;
5580		-}
5581		-
5582		-static int mlx5_ib_read_counters(struct ib_counters *counters,
5583		- struct ib_counters_read_attr *read_attr,
5584		- struct uverbs_attr_bundle *attrs)
5585		-{
5586		- struct mlx5_ib_mcounters *mcounters = to_mcounters(counters);
5587		- struct mlx5_read_counters_attr mread_attr = {};
5588		- struct mlx5_ib_flow_counters_desc *desc;
5589		- int ret, i;
5590		-
5591		- mutex_lock(&mcounters->mcntrs_mutex);
5592		- if (mcounters->cntrs_max_index > read_attr->ncounters) {
5593		- ret = -EINVAL;
5594		- goto err_bound;
5595		- }
5596		-
5597		- mread_attr.out = kcalloc(mcounters->counters_num, sizeof(u64),
5598		- GFP_KERNEL);
5599		- if (!mread_attr.out) {
5600		- ret = -ENOMEM;
5601		- goto err_bound;
5602		- }
5603		-
5604		- mread_attr.hw_cntrs_hndl = mcounters->hw_cntrs_hndl;
5605		- mread_attr.flags = read_attr->flags;
5606		- ret = mcounters->read_counters(counters->device, &mread_attr);
5607		- if (ret)
5608		- goto err_read;
5609		-
5610		- /* do the pass over the counters data array to assign according to the
5611		- * descriptions and indexing pairs
5612		- */
5613		- desc = mcounters->counters_data;
5614		- for (i = 0; i < mcounters->ncounters; i++)
5615		- read_attr->counters_buff[desc[i].index] += mread_attr.out[desc[i].description];
5616		-
5617		-err_read:
5618		- kfree(mread_attr.out);
5619		-err_bound:
5620		- mutex_unlock(&mcounters->mcntrs_mutex);
5621		- return ret;
5622		-}
5623		-
5624		-static int mlx5_ib_destroy_counters(struct ib_counters *counters)
5625		-{
5626		- struct mlx5_ib_mcounters *mcounters = to_mcounters(counters);
5627		-
5628		- counters_clear_description(counters);
5629		- if (mcounters->hw_cntrs_hndl)
5630		- mlx5_fc_destroy(to_mdev(counters->device)->mdev,
5631		- mcounters->hw_cntrs_hndl);
5632		-
5633		- kfree(mcounters);
5634		-
5635		- return 0;
5636		-}
5637		-
5638		-static struct ib_counters mlx5_ib_create_counters(struct ib_device device,
5639		- struct uverbs_attr_bundle *attrs)
5640		-{
5641		- struct mlx5_ib_mcounters *mcounters;
5642		-
5643		- mcounters = kzalloc(sizeof(*mcounters), GFP_KERNEL);
5644		- if (!mcounters)
5645		- return ERR_PTR(-ENOMEM);
5646		-
5647		- mutex_init(&mcounters->mcntrs_mutex);
5648		-
5649		- return &mcounters->ibcntrs;
5650		-}
5651		-
5652		-void mlx5_ib_stage_init_cleanup(struct mlx5_ib_dev *dev)
5653		-{
5654		- mlx5_ib_cleanup_multiport_master(dev);
5655		-#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
5656		- cleanup_srcu_struct(&dev->mr_srcu);
5657		-#endif
5658		- kfree(dev->port);
5659		-}
5660		-
5661		-int mlx5_ib_stage_init_init(struct mlx5_ib_dev *dev)
5662		-{
5663		- struct mlx5_core_dev *mdev = dev->mdev;
5664		- const char *name;
5665		- int err;
5666		- int i;
5667		-
5668		- dev->port = kcalloc(dev->num_ports, sizeof(*dev->port),
5669		- GFP_KERNEL);
5670		- if (!dev->port)
5671		- return -ENOMEM;
5672		-
5673		- for (i = 0; i < dev->num_ports; i++) {
5674		- spin_lock_init(&dev->port[i].mp.mpi_lock);
5675		- rwlock_init(&dev->roce[i].netdev_lock);
5676		- }
5677		-
5678		- err = mlx5_ib_init_multiport_master(dev);
5679		- if (err)
5680		- goto err_free_port;
5681		-
5682		- if (!mlx5_core_mp_enabled(mdev)) {
5683		- for (i = 1; i <= dev->num_ports; i++) {
5684		- err = get_port_caps(dev, i);
5685		- if (err)
5686		- break;
5687		- }
5688		- } else {
5689		- err = get_port_caps(dev, mlx5_core_native_port_num(mdev));
5690		- }
5691		- if (err)
5692		- goto err_mp;
5693		-
5694		- if (mlx5_use_mad_ifc(dev))
5695		- get_ext_port_caps(dev);
5696		-
5697		- if (!mlx5_lag_is_active(mdev))
5698		- name = "mlx5_%d";
5699		- else
5700		- name = "mlx5_bond_%d";
5701		-
5702		- strlcpy(dev->ib_dev.name, name, IB_DEVICE_NAME_MAX);
5703		- dev->ib_dev.owner = THIS_MODULE;
5704		- dev->ib_dev.node_type = RDMA_NODE_IB_CA;
5705		- dev->ib_dev.local_dma_lkey = 0 /* not supported for now */;
5706		- dev->ib_dev.phys_port_cnt = dev->num_ports;
5707		- dev->ib_dev.num_comp_vectors =
5708		- dev->mdev->priv.eq_table.num_comp_vectors;
5709		- dev->ib_dev.dev.parent = &mdev->pdev->dev;
5710		-
5711		- mutex_init(&dev->cap_mask_mutex);
5712		- INIT_LIST_HEAD(&dev->qp_list);
5713		- spin_lock_init(&dev->reset_flow_resource_lock);
5714		-
5715		- spin_lock_init(&dev->memic.memic_lock);
5716		- dev->memic.dev = mdev;
5717		-
5718		-#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
5719		- err = init_srcu_struct(&dev->mr_srcu);
5720		- if (err)
5721		- goto err_free_port;
5722		-#endif
5723		-
5724		- return 0;
5725		-err_mp:
5726		- mlx5_ib_cleanup_multiport_master(dev);
5727		-
5728		-err_free_port:
5729		- kfree(dev->port);
5730		-
5731		- return -ENOMEM;
5732		-}
5733		-
5734		-static int mlx5_ib_stage_flow_db_init(struct mlx5_ib_dev *dev)
5735		-{
5736		- dev->flow_db = kzalloc(sizeof(*dev->flow_db), GFP_KERNEL);
5737		-
5738		- if (!dev->flow_db)
5739		- return -ENOMEM;
5740		-
5741		- mutex_init(&dev->flow_db->lock);
5742		-
5743		- return 0;
5744		-}
5745		-
5746		-int mlx5_ib_stage_rep_flow_db_init(struct mlx5_ib_dev *dev)
5747		-{
5748		- struct mlx5_ib_dev *nic_dev;
5749		-
5750		- nic_dev = mlx5_ib_get_uplink_ibdev(dev->mdev->priv.eswitch);
5751		-
5752		- if (!nic_dev)
5753		- return -EINVAL;
5754		-
5755		- dev->flow_db = nic_dev->flow_db;
5756		-
5757		- return 0;
5758		-}
5759		-
5760		-static void mlx5_ib_stage_flow_db_cleanup(struct mlx5_ib_dev *dev)
5761		-{
5762		- kfree(dev->flow_db);
5763		-}
5764		-
5765		-int mlx5_ib_stage_caps_init(struct mlx5_ib_dev *dev)
5766		-{
5767		- struct mlx5_core_dev *mdev = dev->mdev;
5768		- int err;
5769		-
5770		- dev->ib_dev.uverbs_abi_ver = MLX5_IB_UVERBS_ABI_VERSION;
5771		- dev->ib_dev.uverbs_cmd_mask =
5772		- (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) \|
5773		- (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) \|
5774		- (1ull << IB_USER_VERBS_CMD_QUERY_PORT) \|
5775		- (1ull << IB_USER_VERBS_CMD_ALLOC_PD) \|
5776		- (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) \|
5777		- (1ull << IB_USER_VERBS_CMD_CREATE_AH) \|
5778		- (1ull << IB_USER_VERBS_CMD_DESTROY_AH) \|
5779		- (1ull << IB_USER_VERBS_CMD_REG_MR) \|
5780		- (1ull << IB_USER_VERBS_CMD_REREG_MR) \|
5781		- (1ull << IB_USER_VERBS_CMD_DEREG_MR) \|
5782		- (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) \|
5783		- (1ull << IB_USER_VERBS_CMD_CREATE_CQ) \|
5784		- (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) \|
5785		- (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) \|
5786		- (1ull << IB_USER_VERBS_CMD_CREATE_QP) \|
5787		- (1ull << IB_USER_VERBS_CMD_MODIFY_QP) \|
5788		- (1ull << IB_USER_VERBS_CMD_QUERY_QP) \|
5789		- (1ull << IB_USER_VERBS_CMD_DESTROY_QP) \|
5790		- (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) \|
5791		- (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) \|
5792		- (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) \|
5793		- (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) \|
5794		- (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) \|
5795		- (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) \|
5796		- (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ) \|
5797		- (1ull << IB_USER_VERBS_CMD_OPEN_QP);
5798		- dev->ib_dev.uverbs_ex_cmd_mask =
5799		- (1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) \|
5800		- (1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) \|
5801		- (1ull << IB_USER_VERBS_EX_CMD_CREATE_QP) \|
5802		- (1ull << IB_USER_VERBS_EX_CMD_MODIFY_QP) \|
5803		- (1ull << IB_USER_VERBS_EX_CMD_MODIFY_CQ);
5804		-
5805		- dev->ib_dev.query_device = mlx5_ib_query_device;
5806		- dev->ib_dev.get_link_layer = mlx5_ib_port_link_layer;
5807		- dev->ib_dev.query_gid = mlx5_ib_query_gid;
5808		- dev->ib_dev.add_gid = mlx5_ib_add_gid;
5809		- dev->ib_dev.del_gid = mlx5_ib_del_gid;
5810		- dev->ib_dev.query_pkey = mlx5_ib_query_pkey;
5811		- dev->ib_dev.modify_device = mlx5_ib_modify_device;
5812		- dev->ib_dev.modify_port = mlx5_ib_modify_port;
5813		- dev->ib_dev.alloc_ucontext = mlx5_ib_alloc_ucontext;
5814		- dev->ib_dev.dealloc_ucontext = mlx5_ib_dealloc_ucontext;
5815		- dev->ib_dev.mmap = mlx5_ib_mmap;
5816		- dev->ib_dev.alloc_pd = mlx5_ib_alloc_pd;
5817		- dev->ib_dev.dealloc_pd = mlx5_ib_dealloc_pd;
5818		- dev->ib_dev.create_ah = mlx5_ib_create_ah;
5819		- dev->ib_dev.query_ah = mlx5_ib_query_ah;
5820		- dev->ib_dev.destroy_ah = mlx5_ib_destroy_ah;
5821		- dev->ib_dev.create_srq = mlx5_ib_create_srq;
5822		- dev->ib_dev.modify_srq = mlx5_ib_modify_srq;
5823		- dev->ib_dev.query_srq = mlx5_ib_query_srq;
5824		- dev->ib_dev.destroy_srq = mlx5_ib_destroy_srq;
5825		- dev->ib_dev.post_srq_recv = mlx5_ib_post_srq_recv;
5826		- dev->ib_dev.create_qp = mlx5_ib_create_qp;
5827		- dev->ib_dev.modify_qp = mlx5_ib_modify_qp;
5828		- dev->ib_dev.query_qp = mlx5_ib_query_qp;
5829		- dev->ib_dev.destroy_qp = mlx5_ib_destroy_qp;
5830		- dev->ib_dev.drain_sq = mlx5_ib_drain_sq;
5831		- dev->ib_dev.drain_rq = mlx5_ib_drain_rq;
5832		- dev->ib_dev.post_send = mlx5_ib_post_send;
5833		- dev->ib_dev.post_recv = mlx5_ib_post_recv;
5834		- dev->ib_dev.create_cq = mlx5_ib_create_cq;
5835		- dev->ib_dev.modify_cq = mlx5_ib_modify_cq;
5836		- dev->ib_dev.resize_cq = mlx5_ib_resize_cq;
5837		- dev->ib_dev.destroy_cq = mlx5_ib_destroy_cq;
5838		- dev->ib_dev.poll_cq = mlx5_ib_poll_cq;
5839		- dev->ib_dev.req_notify_cq = mlx5_ib_arm_cq;
5840		- dev->ib_dev.get_dma_mr = mlx5_ib_get_dma_mr;
5841		- dev->ib_dev.reg_user_mr = mlx5_ib_reg_user_mr;
5842		- dev->ib_dev.rereg_user_mr = mlx5_ib_rereg_user_mr;
5843		- dev->ib_dev.dereg_mr = mlx5_ib_dereg_mr;
5844		- dev->ib_dev.attach_mcast = mlx5_ib_mcg_attach;
5845		- dev->ib_dev.detach_mcast = mlx5_ib_mcg_detach;
5846		- dev->ib_dev.process_mad = mlx5_ib_process_mad;
5847		- dev->ib_dev.alloc_mr = mlx5_ib_alloc_mr;
5848		- dev->ib_dev.map_mr_sg = mlx5_ib_map_mr_sg;
5849		- dev->ib_dev.check_mr_status = mlx5_ib_check_mr_status;
5850		- dev->ib_dev.get_dev_fw_str = get_dev_fw_str;
5851		- dev->ib_dev.get_vector_affinity = mlx5_ib_get_vector_affinity;
5852		- if (MLX5_CAP_GEN(mdev, ipoib_enhanced_offloads))
5853		- dev->ib_dev.alloc_rdma_netdev = mlx5_ib_alloc_rdma_netdev;
5854		-
5855		- if (mlx5_core_is_pf(mdev)) {
5856		- dev->ib_dev.get_vf_config = mlx5_ib_get_vf_config;
5857		- dev->ib_dev.set_vf_link_state = mlx5_ib_set_vf_link_state;
5858		- dev->ib_dev.get_vf_stats = mlx5_ib_get_vf_stats;
5859		- dev->ib_dev.set_vf_guid = mlx5_ib_set_vf_guid;
5860		- }
5861		-
5862		- dev->ib_dev.disassociate_ucontext = mlx5_ib_disassociate_ucontext;
5863		-
5864		- dev->umr_fence = mlx5_get_umr_fence(MLX5_CAP_GEN(mdev, umr_fence));
5865		-
5866		- if (MLX5_CAP_GEN(mdev, imaicl)) {
5867		- dev->ib_dev.alloc_mw = mlx5_ib_alloc_mw;
5868		- dev->ib_dev.dealloc_mw = mlx5_ib_dealloc_mw;
5869		- dev->ib_dev.uverbs_cmd_mask \|=
5870		- (1ull << IB_USER_VERBS_CMD_ALLOC_MW) \|
5871		- (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
5872		- }
5873		-
5874		- if (MLX5_CAP_GEN(mdev, xrc)) {
5875		- dev->ib_dev.alloc_xrcd = mlx5_ib_alloc_xrcd;
5876		- dev->ib_dev.dealloc_xrcd = mlx5_ib_dealloc_xrcd;
5877		- dev->ib_dev.uverbs_cmd_mask \|=
5878		- (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) \|
5879		- (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
5880		- }
5881		-
5882		- if (MLX5_CAP_DEV_MEM(mdev, memic)) {
5883		- dev->ib_dev.alloc_dm = mlx5_ib_alloc_dm;
5884		- dev->ib_dev.dealloc_dm = mlx5_ib_dealloc_dm;
5885		- dev->ib_dev.reg_dm_mr = mlx5_ib_reg_dm_mr;
5886		- }
5887		-
5888		- dev->ib_dev.create_flow = mlx5_ib_create_flow;
5889		- dev->ib_dev.destroy_flow = mlx5_ib_destroy_flow;
5890		- dev->ib_dev.uverbs_ex_cmd_mask \|=
5891		- (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) \|
5892		- (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
5893		- dev->ib_dev.create_flow_action_esp = mlx5_ib_create_flow_action_esp;
5894		- dev->ib_dev.destroy_flow_action = mlx5_ib_destroy_flow_action;
5895		- dev->ib_dev.modify_flow_action_esp = mlx5_ib_modify_flow_action_esp;
5896		- dev->ib_dev.driver_id = RDMA_DRIVER_MLX5;
5897		- dev->ib_dev.create_counters = mlx5_ib_create_counters;
5898		- dev->ib_dev.destroy_counters = mlx5_ib_destroy_counters;
5899		- dev->ib_dev.read_counters = mlx5_ib_read_counters;
5900		-
5901		- err = init_node_data(dev);
5902		- if (err)
5903		- return err;
5904		-
5905		- if ((MLX5_CAP_GEN(dev->mdev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
5906		- (MLX5_CAP_GEN(dev->mdev, disable_local_lb_uc) \|\|
5907		- MLX5_CAP_GEN(dev->mdev, disable_local_lb_mc)))
5908		- mutex_init(&dev->lb_mutex);
5909		-
5910		- return 0;
5911		-}
5912		-
5913		-static int mlx5_ib_stage_non_default_cb(struct mlx5_ib_dev *dev)
5914		-{
5915		- dev->ib_dev.get_port_immutable = mlx5_port_immutable;
5916		- dev->ib_dev.query_port = mlx5_ib_query_port;
5917		-
5918		- return 0;
5919		-}
5920		-
5921		-int mlx5_ib_stage_rep_non_default_cb(struct mlx5_ib_dev *dev)
5922		-{
5923		- dev->ib_dev.get_port_immutable = mlx5_port_rep_immutable;
5924		- dev->ib_dev.query_port = mlx5_ib_rep_query_port;
5925		-
5926		- return 0;
5927		-}
5928		-
5929		-static int mlx5_ib_stage_common_roce_init(struct mlx5_ib_dev *dev)
5930		-{
5931		- u8 port_num;
5932		- int i;
5933		-
5934		- for (i = 0; i < dev->num_ports; i++) {
5935		- dev->roce[i].dev = dev;
5936		- dev->roce[i].native_port_num = i + 1;
5937		- dev->roce[i].last_port_state = IB_PORT_DOWN;
5938		- }
5939		-
5940		- dev->ib_dev.get_netdev = mlx5_ib_get_netdev;
5941		- dev->ib_dev.create_wq = mlx5_ib_create_wq;
5942		- dev->ib_dev.modify_wq = mlx5_ib_modify_wq;
5943		- dev->ib_dev.destroy_wq = mlx5_ib_destroy_wq;
5944		- dev->ib_dev.create_rwq_ind_table = mlx5_ib_create_rwq_ind_table;
5945		- dev->ib_dev.destroy_rwq_ind_table = mlx5_ib_destroy_rwq_ind_table;
5946		-
5947		- dev->ib_dev.uverbs_ex_cmd_mask \|=
5948		- (1ull << IB_USER_VERBS_EX_CMD_CREATE_WQ) \|
5949		- (1ull << IB_USER_VERBS_EX_CMD_MODIFY_WQ) \|
5950		- (1ull << IB_USER_VERBS_EX_CMD_DESTROY_WQ) \|
5951		- (1ull << IB_USER_VERBS_EX_CMD_CREATE_RWQ_IND_TBL) \|
5952		- (1ull << IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL);
5953		-
5954		- port_num = mlx5_core_native_port_num(dev->mdev) - 1;
5955		-
5956		- return mlx5_add_netdev_notifier(dev, port_num);
5957		-}
5958		-
5959		-static void mlx5_ib_stage_common_roce_cleanup(struct mlx5_ib_dev *dev)
5960		-{
5961		- u8 port_num = mlx5_core_native_port_num(dev->mdev) - 1;
5962		-
5963		- mlx5_remove_netdev_notifier(dev, port_num);
5964		-}
5965		-
5966		-int mlx5_ib_stage_rep_roce_init(struct mlx5_ib_dev *dev)
5967		-{
5968		- struct mlx5_core_dev *mdev = dev->mdev;
5969		- enum rdma_link_layer ll;
5970		- int port_type_cap;
5971		- int err = 0;
5972		-
5973		- port_type_cap = MLX5_CAP_GEN(mdev, port_type);
5974		- ll = mlx5_port_type_cap_to_rdma_ll(port_type_cap);
5975		-
5976		- if (ll == IB_LINK_LAYER_ETHERNET)
5977		- err = mlx5_ib_stage_common_roce_init(dev);
5978		-
5979		- return err;
5980		-}
5981		-
5982		-void mlx5_ib_stage_rep_roce_cleanup(struct mlx5_ib_dev *dev)
5983		-{
5984		- mlx5_ib_stage_common_roce_cleanup(dev);
5985		-}
5986		-
5987		-static int mlx5_ib_stage_roce_init(struct mlx5_ib_dev *dev)
5988		-{
5989		- struct mlx5_core_dev *mdev = dev->mdev;
5990		- enum rdma_link_layer ll;
5991		- int port_type_cap;
5992		- int err;
5993		-
5994		- port_type_cap = MLX5_CAP_GEN(mdev, port_type);
5995		- ll = mlx5_port_type_cap_to_rdma_ll(port_type_cap);
5996		-
5997		- if (ll == IB_LINK_LAYER_ETHERNET) {
5998		- err = mlx5_ib_stage_common_roce_init(dev);
5999		- if (err)
6000		- return err;
6001		-
6002		- err = mlx5_enable_eth(dev);
6003		- if (err)
6004		- goto cleanup;
6005		- }
6006		-
6007		- return 0;
6008		-cleanup:
6009		- mlx5_ib_stage_common_roce_cleanup(dev);
6010		-
6011		- return err;
6012		-}
6013		-
6014		-static void mlx5_ib_stage_roce_cleanup(struct mlx5_ib_dev *dev)
6015		-{
6016		- struct mlx5_core_dev *mdev = dev->mdev;
6017		- enum rdma_link_layer ll;
6018		- int port_type_cap;
6019		-
6020		- port_type_cap = MLX5_CAP_GEN(mdev, port_type);
6021		- ll = mlx5_port_type_cap_to_rdma_ll(port_type_cap);
6022		-
6023		- if (ll == IB_LINK_LAYER_ETHERNET) {
6024		- mlx5_disable_eth(dev);
6025		- mlx5_ib_stage_common_roce_cleanup(dev);
6026		- }
6027		-}
6028		-
6029		-int mlx5_ib_stage_dev_res_init(struct mlx5_ib_dev *dev)
6030		-{
6031		- return create_dev_resources(&dev->devr);
6032		-}
6033		-
6034		-void mlx5_ib_stage_dev_res_cleanup(struct mlx5_ib_dev *dev)
6035		-{
6036		- destroy_dev_resources(&dev->devr);
6037		-}
6038		-
6039		-static int mlx5_ib_stage_odp_init(struct mlx5_ib_dev *dev)
6040		-{
6041		- mlx5_ib_internal_fill_odp_caps(dev);
6042		-
6043		- return mlx5_ib_odp_init_one(dev);
6044		-}
6045		-
6046		-int mlx5_ib_stage_counters_init(struct mlx5_ib_dev *dev)
6047		-{
6048		- if (MLX5_CAP_GEN(dev->mdev, max_qp_cnt)) {
6049		- dev->ib_dev.get_hw_stats = mlx5_ib_get_hw_stats;
6050		- dev->ib_dev.alloc_hw_stats = mlx5_ib_alloc_hw_stats;
6051		-
6052		- return mlx5_ib_alloc_counters(dev);
6053		- }
6054		-
6055		- return 0;
6056		-}
6057		-
6058		-void mlx5_ib_stage_counters_cleanup(struct mlx5_ib_dev *dev)
6059		-{
6060		- if (MLX5_CAP_GEN(dev->mdev, max_qp_cnt))
6061		- mlx5_ib_dealloc_counters(dev);
6062		-}
6063		-
6064		-static int mlx5_ib_stage_cong_debugfs_init(struct mlx5_ib_dev *dev)
6065		-{
6066		- return mlx5_ib_init_cong_debugfs(dev,
6067		- mlx5_core_native_port_num(dev->mdev) - 1);
6068		-}
6069		-
6070		-static void mlx5_ib_stage_cong_debugfs_cleanup(struct mlx5_ib_dev *dev)
6071		-{
6072		- mlx5_ib_cleanup_cong_debugfs(dev,
6073		- mlx5_core_native_port_num(dev->mdev) - 1);
6074		-}
6075		-
6076		-static int mlx5_ib_stage_uar_init(struct mlx5_ib_dev *dev)
6077		-{
6078		- dev->mdev->priv.uar = mlx5_get_uars_page(dev->mdev);
6079		- return PTR_ERR_OR_ZERO(dev->mdev->priv.uar);
6080		-}
6081		-
6082		-static void mlx5_ib_stage_uar_cleanup(struct mlx5_ib_dev *dev)
6083		-{
6084		- mlx5_put_uars_page(dev->mdev, dev->mdev->priv.uar);
6085		-}
6086		-
6087		-int mlx5_ib_stage_bfrag_init(struct mlx5_ib_dev *dev)
6088		-{
6089		- int err;
6090		-
6091		- err = mlx5_alloc_bfreg(dev->mdev, &dev->bfreg, false, false);
6092		- if (err)
6093		- return err;
6094		-
6095		- err = mlx5_alloc_bfreg(dev->mdev, &dev->fp_bfreg, false, true);
6096		- if (err)
6097		- mlx5_free_bfreg(dev->mdev, &dev->bfreg);
6098		-
6099		- return err;
6100		-}
6101		-
6102		-void mlx5_ib_stage_bfrag_cleanup(struct mlx5_ib_dev *dev)
6103		-{
6104		- mlx5_free_bfreg(dev->mdev, &dev->fp_bfreg);
6105		- mlx5_free_bfreg(dev->mdev, &dev->bfreg);
6106		-}
6107		-
6108		-static int mlx5_ib_stage_populate_specs(struct mlx5_ib_dev *dev)
6109		-{
6110		- return populate_specs_root(dev);
6111		-}
6112		-
6113		-int mlx5_ib_stage_ib_reg_init(struct mlx5_ib_dev *dev)
6114		-{
6115		- return ib_register_device(&dev->ib_dev, NULL);
6116		-}
6117		-
6118		-void mlx5_ib_stage_pre_ib_reg_umr_cleanup(struct mlx5_ib_dev *dev)
6119		-{
6120		- destroy_umrc_res(dev);
6121		-}
6122		-
6123		-void mlx5_ib_stage_ib_reg_cleanup(struct mlx5_ib_dev *dev)
6124		-{
6125		- ib_unregister_device(&dev->ib_dev);
6126		-}
6127		-
6128		-int mlx5_ib_stage_post_ib_reg_umr_init(struct mlx5_ib_dev *dev)
6129		-{
6130		- return create_umr_res(dev);
6131		-}
6132		-
6133		-static int mlx5_ib_stage_delay_drop_init(struct mlx5_ib_dev *dev)
6134		-{
6135		- init_delay_drop(dev);
6136		-
	4547	+ debugfs_create_atomic_t("num_timeout_events", 0400, root,
	4548	+ &dev->delay_drop.events_cnt);
	4549	+ debugfs_create_atomic_t("num_rqs", 0400, root,
	4550	+ &dev->delay_drop.rqs_cnt);
	4551	+ debugfs_create_file("timeout", 0600, root, &dev->delay_drop,
	4552	+ &fops_delay_drop_timeout);
6137	4553	return 0;
6138	4554	}
6139	4555
6140	4556	static void mlx5_ib_stage_delay_drop_cleanup(struct mlx5_ib_dev *dev)
6141	4557	{
6142		- cancel_delay_drop(dev);
	4558	+ if (!(dev->ib_dev.attrs.raw_packet_caps & IB_RAW_PACKET_CAP_DELAY_DROP))
	4559	+ return;
	4560	+
	4561	+ cancel_work_sync(&dev->delay_drop.delay_drop_work);
	4562	+ if (!dev->delay_drop.dir_debugfs)
	4563	+ return;
	4564	+
	4565	+ debugfs_remove_recursive(dev->delay_drop.dir_debugfs);
	4566	+ dev->delay_drop.dir_debugfs = NULL;
6143	4567	}
6144	4568
6145		-int mlx5_ib_stage_class_attr_init(struct mlx5_ib_dev *dev)
	4569	+static int mlx5_ib_stage_dev_notifier_init(struct mlx5_ib_dev *dev)
6146	4570	{
6147		- int err;
6148		- int i;
6149		-
6150		- for (i = 0; i < ARRAY_SIZE(mlx5_class_attributes); i++) {
6151		- err = device_create_file(&dev->ib_dev.dev,
6152		- mlx5_class_attributes[i]);
6153		- if (err)
6154		- return err;
6155		- }
6156		-
	4571	+ dev->mdev_events.notifier_call = mlx5_ib_event;
	4572	+ mlx5_notifier_register(dev->mdev, &dev->mdev_events);
6157	4573	return 0;
6158	4574	}
6159	4575
6160		-static int mlx5_ib_stage_rep_reg_init(struct mlx5_ib_dev *dev)
	4576	+static void mlx5_ib_stage_dev_notifier_cleanup(struct mlx5_ib_dev *dev)
6161	4577	{
6162		- mlx5_ib_register_vport_reps(dev);
6163		-
6164		- return 0;
6165		-}
6166		-
6167		-static void mlx5_ib_stage_rep_reg_cleanup(struct mlx5_ib_dev *dev)
6168		-{
6169		- mlx5_ib_unregister_vport_reps(dev);
	4578	+ mlx5_notifier_unregister(dev->mdev, &dev->mdev_events);
6170	4579	}
6171	4580
6172	4581	void __mlx5_ib_remove(struct mlx5_ib_dev *dev,
6173	4582	const struct mlx5_ib_profile *profile,
6174	4583	int stage)
6175	4584	{
	4585	+ dev->ib_active = false;
	4586	+
6176	4587	/* Number of stages to cleanup */
6177	4588	while (stage) {
6178	4589	stage--;
..	..	@@ -6180,7 +4591,8 @@
6180	4591	profile->stage[stage].cleanup(dev);
6181	4592	}
6182	4593
6183		- ib_dealloc_device((struct ib_device *)dev);
	4594	+ kfree(dev->port);
	4595	+ ib_dealloc_device(&dev->ib_dev);
6184	4596	}
6185	4597
6186	4598	void __mlx5_ib_add(struct mlx5_ib_dev dev,
..	..	@@ -6189,7 +4601,7 @@
6189	4601	int err;
6190	4602	int i;
6191	4603
6192		- printk_once(KERN_INFO "%s", mlx5_version);
	4604	+ dev->profile = profile;
6193	4605
6194	4606	for (i = 0; i < MLX5_IB_STAGE_MAX; i++) {
6195	4607	if (profile->stage[i].init) {
..	..	@@ -6199,7 +4611,6 @@
6199	4611	}
6200	4612	}
6201	4613
6202		- dev->profile = profile;
6203	4614	dev->ib_active = true;
6204	4615
6205	4616	return dev;
..	..	@@ -6214,27 +4625,36 @@
6214	4625	STAGE_CREATE(MLX5_IB_STAGE_INIT,
6215	4626	mlx5_ib_stage_init_init,
6216	4627	mlx5_ib_stage_init_cleanup),
6217		- STAGE_CREATE(MLX5_IB_STAGE_FLOW_DB,
6218		- mlx5_ib_stage_flow_db_init,
6219		- mlx5_ib_stage_flow_db_cleanup),
	4628	+ STAGE_CREATE(MLX5_IB_STAGE_FS,
	4629	+ mlx5_ib_fs_init,
	4630	+ mlx5_ib_fs_cleanup),
6220	4631	STAGE_CREATE(MLX5_IB_STAGE_CAPS,
6221	4632	mlx5_ib_stage_caps_init,
6222		- NULL),
	4633	+ mlx5_ib_stage_caps_cleanup),
6223	4634	STAGE_CREATE(MLX5_IB_STAGE_NON_DEFAULT_CB,
6224	4635	mlx5_ib_stage_non_default_cb,
6225	4636	NULL),
6226	4637	STAGE_CREATE(MLX5_IB_STAGE_ROCE,
6227		- mlx5_ib_stage_roce_init,
6228		- mlx5_ib_stage_roce_cleanup),
	4638	+ mlx5_ib_roce_init,
	4639	+ mlx5_ib_roce_cleanup),
	4640	+ STAGE_CREATE(MLX5_IB_STAGE_QP,
	4641	+ mlx5_init_qp_table,
	4642	+ mlx5_cleanup_qp_table),
	4643	+ STAGE_CREATE(MLX5_IB_STAGE_SRQ,
	4644	+ mlx5_init_srq_table,
	4645	+ mlx5_cleanup_srq_table),
6229	4646	STAGE_CREATE(MLX5_IB_STAGE_DEVICE_RESOURCES,
6230		- mlx5_ib_stage_dev_res_init,
6231		- mlx5_ib_stage_dev_res_cleanup),
	4647	+ mlx5_ib_dev_res_init,
	4648	+ mlx5_ib_dev_res_cleanup),
	4649	+ STAGE_CREATE(MLX5_IB_STAGE_DEVICE_NOTIFIER,
	4650	+ mlx5_ib_stage_dev_notifier_init,
	4651	+ mlx5_ib_stage_dev_notifier_cleanup),
6232	4652	STAGE_CREATE(MLX5_IB_STAGE_ODP,
6233		- mlx5_ib_stage_odp_init,
6234		- NULL),
	4653	+ mlx5_ib_odp_init_one,
	4654	+ mlx5_ib_odp_cleanup_one),
6235	4655	STAGE_CREATE(MLX5_IB_STAGE_COUNTERS,
6236		- mlx5_ib_stage_counters_init,
6237		- mlx5_ib_stage_counters_cleanup),
	4656	+ mlx5_ib_counters_init,
	4657	+ mlx5_ib_counters_cleanup),
6238	4658	STAGE_CREATE(MLX5_IB_STAGE_CONG_DEBUGFS,
6239	4659	mlx5_ib_stage_cong_debugfs_init,
6240	4660	mlx5_ib_stage_cong_debugfs_cleanup),
..	..	@@ -6247,9 +4667,9 @@
6247	4667	STAGE_CREATE(MLX5_IB_STAGE_PRE_IB_REG_UMR,
6248	4668	NULL,
6249	4669	mlx5_ib_stage_pre_ib_reg_umr_cleanup),
6250		- STAGE_CREATE(MLX5_IB_STAGE_SPECS,
6251		- mlx5_ib_stage_populate_specs,
6252		- NULL),
	4670	+ STAGE_CREATE(MLX5_IB_STAGE_WHITELIST_UID,
	4671	+ mlx5_ib_devx_init,
	4672	+ mlx5_ib_devx_cleanup),
6253	4673	STAGE_CREATE(MLX5_IB_STAGE_IB_REG,
6254	4674	mlx5_ib_stage_ib_reg_init,
6255	4675	mlx5_ib_stage_ib_reg_cleanup),
..	..	@@ -6259,33 +4679,45 @@
6259	4679	STAGE_CREATE(MLX5_IB_STAGE_DELAY_DROP,
6260	4680	mlx5_ib_stage_delay_drop_init,
6261	4681	mlx5_ib_stage_delay_drop_cleanup),
6262		- STAGE_CREATE(MLX5_IB_STAGE_CLASS_ATTR,
6263		- mlx5_ib_stage_class_attr_init,
	4682	+ STAGE_CREATE(MLX5_IB_STAGE_RESTRACK,
	4683	+ mlx5_ib_restrack_init,
6264	4684	NULL),
6265	4685	};
6266	4686
6267		-static const struct mlx5_ib_profile nic_rep_profile = {
	4687	+const struct mlx5_ib_profile raw_eth_profile = {
6268	4688	STAGE_CREATE(MLX5_IB_STAGE_INIT,
6269	4689	mlx5_ib_stage_init_init,
6270	4690	mlx5_ib_stage_init_cleanup),
6271		- STAGE_CREATE(MLX5_IB_STAGE_FLOW_DB,
6272		- mlx5_ib_stage_flow_db_init,
6273		- mlx5_ib_stage_flow_db_cleanup),
	4691	+ STAGE_CREATE(MLX5_IB_STAGE_FS,
	4692	+ mlx5_ib_fs_init,
	4693	+ mlx5_ib_fs_cleanup),
6274	4694	STAGE_CREATE(MLX5_IB_STAGE_CAPS,
6275	4695	mlx5_ib_stage_caps_init,
6276		- NULL),
	4696	+ mlx5_ib_stage_caps_cleanup),
6277	4697	STAGE_CREATE(MLX5_IB_STAGE_NON_DEFAULT_CB,
6278		- mlx5_ib_stage_rep_non_default_cb,
	4698	+ mlx5_ib_stage_raw_eth_non_default_cb,
6279	4699	NULL),
6280	4700	STAGE_CREATE(MLX5_IB_STAGE_ROCE,
6281		- mlx5_ib_stage_rep_roce_init,
6282		- mlx5_ib_stage_rep_roce_cleanup),
	4701	+ mlx5_ib_roce_init,
	4702	+ mlx5_ib_roce_cleanup),
	4703	+ STAGE_CREATE(MLX5_IB_STAGE_QP,
	4704	+ mlx5_init_qp_table,
	4705	+ mlx5_cleanup_qp_table),
	4706	+ STAGE_CREATE(MLX5_IB_STAGE_SRQ,
	4707	+ mlx5_init_srq_table,
	4708	+ mlx5_cleanup_srq_table),
6283	4709	STAGE_CREATE(MLX5_IB_STAGE_DEVICE_RESOURCES,
6284		- mlx5_ib_stage_dev_res_init,
6285		- mlx5_ib_stage_dev_res_cleanup),
	4710	+ mlx5_ib_dev_res_init,
	4711	+ mlx5_ib_dev_res_cleanup),
	4712	+ STAGE_CREATE(MLX5_IB_STAGE_DEVICE_NOTIFIER,
	4713	+ mlx5_ib_stage_dev_notifier_init,
	4714	+ mlx5_ib_stage_dev_notifier_cleanup),
6286	4715	STAGE_CREATE(MLX5_IB_STAGE_COUNTERS,
6287		- mlx5_ib_stage_counters_init,
6288		- mlx5_ib_stage_counters_cleanup),
	4716	+ mlx5_ib_counters_init,
	4717	+ mlx5_ib_counters_cleanup),
	4718	+ STAGE_CREATE(MLX5_IB_STAGE_CONG_DEBUGFS,
	4719	+ mlx5_ib_stage_cong_debugfs_init,
	4720	+ mlx5_ib_stage_cong_debugfs_cleanup),
6289	4721	STAGE_CREATE(MLX5_IB_STAGE_UAR,
6290	4722	mlx5_ib_stage_uar_init,
6291	4723	mlx5_ib_stage_uar_cleanup),
..	..	@@ -6295,21 +4727,21 @@
6295	4727	STAGE_CREATE(MLX5_IB_STAGE_PRE_IB_REG_UMR,
6296	4728	NULL,
6297	4729	mlx5_ib_stage_pre_ib_reg_umr_cleanup),
6298		- STAGE_CREATE(MLX5_IB_STAGE_SPECS,
6299		- mlx5_ib_stage_populate_specs,
6300		- NULL),
	4730	+ STAGE_CREATE(MLX5_IB_STAGE_WHITELIST_UID,
	4731	+ mlx5_ib_devx_init,
	4732	+ mlx5_ib_devx_cleanup),
6301	4733	STAGE_CREATE(MLX5_IB_STAGE_IB_REG,
6302	4734	mlx5_ib_stage_ib_reg_init,
6303	4735	mlx5_ib_stage_ib_reg_cleanup),
6304	4736	STAGE_CREATE(MLX5_IB_STAGE_POST_IB_REG_UMR,
6305	4737	mlx5_ib_stage_post_ib_reg_umr_init,
6306	4738	NULL),
6307		- STAGE_CREATE(MLX5_IB_STAGE_CLASS_ATTR,
6308		- mlx5_ib_stage_class_attr_init,
	4739	+ STAGE_CREATE(MLX5_IB_STAGE_DELAY_DROP,
	4740	+ mlx5_ib_stage_delay_drop_init,
	4741	+ mlx5_ib_stage_delay_drop_cleanup),
	4742	+ STAGE_CREATE(MLX5_IB_STAGE_RESTRACK,
	4743	+ mlx5_ib_restrack_init,
6309	4744	NULL),
6310		- STAGE_CREATE(MLX5_IB_STAGE_REP_REG,
6311		- mlx5_ib_stage_rep_reg_init,
6312		- mlx5_ib_stage_rep_reg_cleanup),
6313	4745	};
6314	4746
6315	4747	static void mlx5_ib_add_slave_port(struct mlx5_core_dev mdev)
..	..	@@ -6346,7 +4778,8 @@
6346	4778
6347	4779	if (!bound) {
6348	4780	list_add_tail(&mpi->list, &mlx5_ib_unaffiliated_port_list);
6349		- dev_dbg(&mdev->pdev->dev, "no suitable IB device found to bind to, added to unaffiliated list.\n");
	4781	+ dev_dbg(mdev->device,
	4782	+ "no suitable IB device found to bind to, added to unaffiliated list.\n");
6350	4783	}
6351	4784	mutex_unlock(&mlx5_ib_multiport_mutex);
6352	4785
..	..	@@ -6355,11 +4788,18 @@
6355	4788
6356	4789	static void mlx5_ib_add(struct mlx5_core_dev mdev)
6357	4790	{
	4791	+ const struct mlx5_ib_profile *profile;
6358	4792	enum rdma_link_layer ll;
6359	4793	struct mlx5_ib_dev *dev;
6360	4794	int port_type_cap;
	4795	+ int num_ports;
6361	4796
6362		- printk_once(KERN_INFO "%s", mlx5_version);
	4797	+ if (MLX5_ESWITCH_MANAGER(mdev) &&
	4798	+ mlx5_ib_eswitch_mode(mdev->priv.eswitch) == MLX5_ESWITCH_OFFLOADS) {
	4799	+ if (!mlx5_core_mp_enabled(mdev))
	4800	+ mlx5_ib_register_vport_reps(mdev);
	4801	+ return mdev;
	4802	+ }
6363	4803
6364	4804	port_type_cap = MLX5_CAP_GEN(mdev, port_type);
6365	4805	ll = mlx5_port_type_cap_to_rdma_ll(port_type_cap);
..	..	@@ -6367,28 +4807,38 @@
6367	4807	if (mlx5_core_is_mp_slave(mdev) && ll == IB_LINK_LAYER_ETHERNET)
6368	4808	return mlx5_ib_add_slave_port(mdev);
6369	4809
6370		- dev = (struct mlx5_ib_dev )ib_alloc_device(sizeof(dev));
	4810	+ num_ports = max(MLX5_CAP_GEN(mdev, num_ports),
	4811	+ MLX5_CAP_GEN(mdev, num_vhca_ports));
	4812	+ dev = ib_alloc_device(mlx5_ib_dev, ib_dev);
6371	4813	if (!dev)
6372	4814	return NULL;
6373		-
6374		- dev->mdev = mdev;
6375		- dev->num_ports = max(MLX5_CAP_GEN(mdev, num_ports),
6376		- MLX5_CAP_GEN(mdev, num_vhca_ports));
6377		-
6378		- if (MLX5_ESWITCH_MANAGER(mdev) &&
6379		- mlx5_ib_eswitch_mode(mdev->priv.eswitch) == SRIOV_OFFLOADS) {
6380		- dev->rep = mlx5_ib_vport_rep(mdev->priv.eswitch, 0);
6381		-
6382		- return __mlx5_ib_add(dev, &nic_rep_profile);
	4815	+ dev->port = kcalloc(num_ports, sizeof(*dev->port),
	4816	+ GFP_KERNEL);
	4817	+ if (!dev->port) {
	4818	+ ib_dealloc_device(&dev->ib_dev);
	4819	+ return NULL;
6383	4820	}
6384	4821
6385		- return __mlx5_ib_add(dev, &pf_profile);
	4822	+ dev->mdev = mdev;
	4823	+ dev->num_ports = num_ports;
	4824	+
	4825	+ if (ll == IB_LINK_LAYER_ETHERNET && !mlx5_is_roce_enabled(mdev))
	4826	+ profile = &raw_eth_profile;
	4827	+ else
	4828	+ profile = &pf_profile;
	4829	+
	4830	+ return __mlx5_ib_add(dev, profile);
6386	4831	}
6387	4832
6388	4833	static void mlx5_ib_remove(struct mlx5_core_dev mdev, void context)
6389	4834	{
6390	4835	struct mlx5_ib_multiport_info *mpi;
6391	4836	struct mlx5_ib_dev *dev;
	4837	+
	4838	+ if (MLX5_ESWITCH_MANAGER(mdev) && context == mdev) {
	4839	+ mlx5_ib_unregister_vport_reps(mdev);
	4840	+ return;
	4841	+ }
6392	4842
6393	4843	if (mlx5_core_is_mp_slave(mdev)) {
6394	4844	mpi = context;
..	..	@@ -6408,10 +4858,6 @@
6408	4858	static struct mlx5_interface mlx5_ib_interface = {
6409	4859	.add = mlx5_ib_add,
6410	4860	.remove = mlx5_ib_remove,
6411		- .event = mlx5_ib_event,
6412		-#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
6413		- .pfault = mlx5_ib_pfault,
6414		-#endif
6415	4861	.protocol = MLX5_INTERFACE_PROTOCOL_IB,
6416	4862	};
6417	4863