irdma(4): Upgrade the driver to 1.1.11-k

Summary of changes: - postpone mtu size assignment during load to avoid race condition - refactor some of the debug prints - add request reset handler - refactor flush scheduler to increase efficiency and avoid racing - put correct vlan_tag for UD traffic with PFC - suspend QP before going to ERROR state to avoid CQP timout - fix arithmetic error on irdma_debug_bugf - allow debug flag to be settable during driver load - introduce meaningful default values for DCQCN algorithm - interrupt naming convention improvements - skip unsignaled completions in poll_cmpl Signed-off-by: Bartosz Sobczak [email protected] Signed-off-by: Eric Joyner <[email protected]> Reviewed by: hselasky@ MFC after: 1 week Sponsored by: Intel Corporation Differential Revision: https://reviews.freebsd.org/D39173
jargonTSD · Mar 28, 2023 · 3510590 · 3510590
1 parent b674303
commit 3510590
Show file tree

Hide file tree

Showing 37 changed files with 994 additions and 641 deletions.
diff --git a/contrib/ofed/libirdma/abi.h b/contrib/ofed/libirdma/abi.h
@@ -115,6 +115,7 @@ struct irdma_get_context {
 	__u32 rsvd32;
 	__u8 userspace_ver;
 	__u8 rsvd8[3];
+	__aligned_u64 comp_mask;
 
 };
 struct irdma_get_context_resp {
@@ -136,6 +137,7 @@ struct irdma_get_context_resp {
 	__u16 max_hw_sq_chunk;
 	__u8 hw_rev;
 	__u8 rsvd2;
+	__aligned_u64 comp_mask;
 
 };
 struct irdma_ureg_mr {

diff --git a/contrib/ofed/libirdma/i40iw_hw.h b/contrib/ofed/libirdma/i40iw_hw.h
@@ -41,8 +41,8 @@ enum i40iw_device_caps_const {
 	I40IW_MAX_SGE_RD			= 1,
 	I40IW_MAX_PUSH_PAGE_COUNT		= 0,
 	I40IW_MAX_INLINE_DATA_SIZE		= 48,
-	I40IW_MAX_IRD_SIZE			= 63,
-	I40IW_MAX_ORD_SIZE			= 127,
+	I40IW_MAX_IRD_SIZE			= 64,
+	I40IW_MAX_ORD_SIZE			= 64,
 	I40IW_MAX_WQ_ENTRIES			= 2048,
 	I40IW_MAX_WQE_SIZE_RQ			= 128,
 	I40IW_MAX_PDS				= 32768,

diff --git a/contrib/ofed/libirdma/irdma-abi.h b/contrib/ofed/libirdma/irdma-abi.h
@@ -2,7 +2,7 @@
  * SPDX-License-Identifier: (GPL-2.0 WITH Linux-syscall-note) OR Linux-OpenIB)
  *
  *
- * Copyright (c) 2006 - 2021 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2006 - 2022 Intel Corporation.  All rights reserved.
  * Copyright (c) 2005 Topspin Communications.  All rights reserved.
  * Copyright (c) 2005 Cisco Systems.  All rights reserved.
  * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved.
@@ -53,10 +53,15 @@ enum irdma_memreg_type {
 	IRDMA_MEMREG_TYPE_CQ   = 2,
 };
 
+enum {
+	IRDMA_ALLOC_UCTX_USE_RAW_ATTR = 1 << 0,
+};
+
 struct irdma_alloc_ucontext_req {
 	__u32 rsvd32;
 	__u8 userspace_ver;
 	__u8 rsvd8[3];
+	__aligned_u64 comp_mask;
 };
 
 struct irdma_alloc_ucontext_resp {
@@ -77,6 +82,7 @@ struct irdma_alloc_ucontext_resp {
 	__u16 max_hw_sq_chunk;
 	__u8 hw_rev;
 	__u8 rsvd2;
+	__aligned_u64 comp_mask;
 };
 
 struct irdma_alloc_pd_resp {

diff --git a/contrib/ofed/libirdma/irdma_defs.h b/contrib/ofed/libirdma/irdma_defs.h
@@ -75,6 +75,7 @@
 #define IRDMA_CQE_QTYPE_RQ	0
 #define IRDMA_CQE_QTYPE_SQ	1
 
+#define IRDMA_QP_SW_MIN_WQSIZE	8 /* in WRs*/
 #define IRDMA_QP_WQE_MIN_SIZE	32
 #define IRDMA_QP_WQE_MAX_SIZE	256
 #define IRDMA_QP_WQE_MIN_QUANTA 1
@@ -304,6 +305,17 @@
 #define IRDMAPFINT_OICR_PE_PUSH_M BIT(27)
 #define IRDMAPFINT_OICR_PE_CRITERR_M BIT(28)
 
+#define IRDMA_GET_RING_OFFSET(_ring, _i) \
+	( \
+		((_ring).head + (_i)) % (_ring).size \
+	)
+
+#define IRDMA_GET_CQ_ELEM_AT_OFFSET(_cq, _i, _cqe) \
+	{ \
+		register __u32 offset; \
+		offset = IRDMA_GET_RING_OFFSET((_cq)->cq_ring, _i); \
+		(_cqe) = (_cq)->cq_base[offset].buf; \
+	}
 #define IRDMA_GET_CURRENT_CQ_ELEM(_cq) \
 	( \
 		(_cq)->cq_base[IRDMA_RING_CURRENT_HEAD((_cq)->cq_ring)].buf  \
@@ -437,12 +449,6 @@
 		IRDMA_RING_MOVE_HEAD(_ring, _retcode); \
 	}
 
-enum irdma_protocol_used {
-	IRDMA_ANY_PROTOCOL = 0,
-	IRDMA_IWARP_PROTOCOL_ONLY = 1,
-	IRDMA_ROCE_PROTOCOL_ONLY = 2,
-};
-
 enum irdma_qp_wqe_size {
 	IRDMA_WQE_SIZE_32  = 32,
 	IRDMA_WQE_SIZE_64  = 64,

diff --git a/contrib/ofed/libirdma/irdma_uk.c b/contrib/ofed/libirdma/irdma_uk.c
@@ -641,7 +641,7 @@ irdma_copy_inline_data_gen_1(u8 *wqe, struct irdma_sge *sge_list,
 			sge_len -= bytes_copied;
 
 			if (!quanta_bytes_remaining) {
-				/* Remaining inline bytes reside after the hdr */
+				/* Remaining inline bytes reside after hdr */
 				wqe += 16;
 				quanta_bytes_remaining = 32;
 			}
@@ -710,7 +710,7 @@ irdma_copy_inline_data(u8 *wqe, struct irdma_sge *sge_list, u32 num_sges,
 			if (!quanta_bytes_remaining) {
 				quanta_bytes_remaining = 31;
 
-				/* Remaining inline bytes reside after the hdr */
+				/* Remaining inline bytes reside after hdr */
 				if (first_quanta) {
 					first_quanta = false;
 					wqe += 16;
@@ -1111,7 +1111,6 @@ irdma_uk_cq_request_notification(struct irdma_cq_uk *cq,
 	u8 arm_next = 0;
 	u8 arm_seq_num;
 
-	cq->armed = true;
 	get_64bit_val(cq->shadow_area, IRDMA_BYTE_32, &temp_val);
 	arm_seq_num = (u8)FIELD_GET(IRDMA_CQ_DBSA_ARM_SEQ_NUM, temp_val);
 	arm_seq_num++;
@@ -1338,6 +1337,8 @@ irdma_uk_cq_poll_cmpl(struct irdma_cq_uk *cq,
 	info->error = (bool)FIELD_GET(IRDMA_CQ_ERROR, qword3);
 	info->push_dropped = (bool)FIELD_GET(IRDMACQ_PSHDROP, qword3);
 	info->ipv4 = (bool)FIELD_GET(IRDMACQ_IPV4, qword3);
+	get_64bit_val(cqe, IRDMA_BYTE_8, &comp_ctx);
+	qp = (struct irdma_qp_uk *)(irdma_uintptr) comp_ctx;
 	if (info->error) {
 		info->major_err = FIELD_GET(IRDMA_CQ_MAJERR, qword3);
 		info->minor_err = FIELD_GET(IRDMA_CQ_MINERR, qword3);
@@ -1366,10 +1367,7 @@ irdma_uk_cq_poll_cmpl(struct irdma_cq_uk *cq,
 	info->qp_id = (u32)FIELD_GET(IRDMACQ_QPID, qword2);
 	info->ud_src_qpn = (u32)FIELD_GET(IRDMACQ_UDSRCQPN, qword2);
 
-	get_64bit_val(cqe, IRDMA_BYTE_8, &comp_ctx);
-
 	info->solicited_event = (bool)FIELD_GET(IRDMACQ_SOEVENT, qword3);
-	qp = (struct irdma_qp_uk *)(irdma_uintptr) comp_ctx;
 	if (!qp || qp->destroy_pending) {
 		ret_code = EFAULT;
 		goto exit;
@@ -1493,7 +1491,8 @@ irdma_uk_cq_poll_cmpl(struct irdma_cq_uk *cq,
 				}
 			} while (1);
 
-			if (info->op_type == IRDMA_OP_TYPE_BIND_MW && info->minor_err == FLUSH_PROT_ERR)
+			if (info->op_type == IRDMA_OP_TYPE_BIND_MW &&
+			    info->minor_err == FLUSH_PROT_ERR)
 				info->minor_err = FLUSH_MW_BIND_ERR;
 			qp->sq_flush_seen = true;
 			if (!IRDMA_RING_MORE_WORK(qp->sq_ring))

diff --git a/contrib/ofed/libirdma/irdma_umain.c b/contrib/ofed/libirdma/irdma_umain.c
@@ -49,7 +49,7 @@
 /**
  *  Driver version
  */
-char libirdma_version[] = "1.1.5-k";
+char libirdma_version[] = "1.1.11-k";
 
 unsigned int irdma_dbg;
 
@@ -170,7 +170,7 @@ irdma_init_context(struct verbs_device *vdev,
 	iwvctx->uk_attrs.max_hw_sq_chunk = resp.max_hw_sq_chunk;
 	iwvctx->uk_attrs.max_hw_cq_size = resp.max_hw_cq_size;
 	iwvctx->uk_attrs.min_hw_cq_size = resp.min_hw_cq_size;
-	iwvctx->uk_attrs.min_hw_wq_size = IRDMA_MIN_WQ_SIZE_GEN2;
+	iwvctx->uk_attrs.min_hw_wq_size = IRDMA_QP_SW_MIN_WQSIZE;
 	iwvctx->abi_ver = IRDMA_ABI_VER;
 	mmap_key = resp.db_mmap_key;
 

diff --git a/contrib/ofed/libirdma/irdma_umain.h b/contrib/ofed/libirdma/irdma_umain.h
@@ -87,7 +87,8 @@ struct irdma_uvcontext {
 	struct irdma_uk_attrs uk_attrs;
 	void *db;
 	int abi_ver;
-	bool legacy_mode;
+	bool legacy_mode:1;
+	bool use_raw_attrs:1;
 };
 
 struct irdma_uqp;

diff --git a/contrib/ofed/libirdma/irdma_user.h b/contrib/ofed/libirdma/irdma_user.h
@@ -211,7 +211,6 @@ enum irdma_device_caps_const {
 	IRDMA_Q2_BUF_SIZE =			256,
 	IRDMA_QP_CTX_SIZE =			256,
 	IRDMA_MAX_PDS =				262144,
-	IRDMA_MIN_WQ_SIZE_GEN2 =		8,
 };
 
 enum irdma_addressing_type {
@@ -521,7 +520,6 @@ struct irdma_cq_uk {
 	u32 cq_size;
 	struct irdma_ring cq_ring;
 	u8 polarity;
-	bool armed:1;
 	bool avoid_mem_cflct:1;
 };
 

diff --git a/contrib/ofed/libirdma/irdma_uverbs.c b/contrib/ofed/libirdma/irdma_uverbs.c
@@ -1566,11 +1566,10 @@ irdma_ucreate_qp(struct ibv_pd *pd,
 	info.sq_size = info.sq_depth >> info.sq_shift;
 	info.rq_size = info.rq_depth >> info.rq_shift;
 	/**
-	 * For older ABI version (less than 6) passes raw sq and rq
-	 * quanta in cap.max_send_wr and cap.max_recv_wr.
-	 * But then kernel had no way of calculating the actual qp size.
+	 * Maintain backward compatibility with older ABI which pass sq
+	 * and rq depth (in quanta) in cap.max_send_wr a cap.max_recv_wr
 	 */
-	if (iwvctx->abi_ver <= 5) {
+	if (!iwvctx->use_raw_attrs) {
 		attr->cap.max_send_wr = info.sq_size;
 		attr->cap.max_recv_wr = info.rq_size;
 	}

diff --git a/contrib/ofed/libirdma/osdep.h b/contrib/ofed/libirdma/osdep.h
@@ -119,7 +119,7 @@ do {													\
 	irdma_debug(dev, mask, "%s\n", desc);								\
 	irdma_debug(dev, mask, "starting address virt=%p phy=%lxh\n", buf, irdma_get_virt_to_phy(buf));	\
 	for (i = 0; i < size ; i += 8)									\
-		irdma_debug(dev, mask, "index %03d val: %016lx\n", i, ((unsigned long *)buf)[i / 8]);	\
+		irdma_debug(dev, mask, "index %03d val: %016lx\n", i, ((unsigned long *)(buf))[i / 8]);	\
 } while(0)
 
 #define irdma_debug(h, m, s, ...)					\
@@ -137,11 +137,12 @@ do {                                                                    \
         if (irdma_dbg)                                                  \
                 printf("libirdma-%s: " fmt, __func__, ##args); \
 } while (0)
-#define irdma_dev_err(a, b, ...) printf(b, ##__VA_ARGS__)
-#define irdma_dev_warn(a, b, ...) printf(b, ##__VA_ARGS__) /*dev_warn(a, b)*/
+#define irdma_dev_err(ibdev, fmt, ...) \
+	pr_err("%s:%s:%d ERR "fmt, (ibdev)->name, __func__, __LINE__, ##__VA_ARGS__)
+#define irdma_dev_warn(ibdev, fmt, ...) \
+	pr_warn("%s:%s:%d WARN "fmt, (ibdev)->name, __func__, __LINE__, ##__VA_ARGS__)
 #define irdma_dev_info(a, b, ...) printf(b, ##__VA_ARGS__)
 #define irdma_pr_warn printf
-#define ibdev_err(ibdev, fmt, ...)  printf("%s:"fmt, (ibdev)->name, ##__VA_ARGS__)
 
 #define dump_struct(s, sz, name)	\
 do {				\

diff --git a/sys/dev/irdma/fbsd_kcompat.c b/sys/dev/irdma/fbsd_kcompat.c
@@ -93,6 +93,18 @@ irdma_wr64(struct irdma_dev_ctx *dev_ctx, u32 reg, u64 value)
 
 }
 
+void
+irdma_request_reset(struct irdma_pci_f *rf)
+{
+	struct ice_rdma_peer *peer = rf->peer_info;
+	struct ice_rdma_request req = {0};
+
+	req.type = ICE_RDMA_EVENT_RESET;
+
+	printf("%s:%d requesting pf-reset\n", __func__, __LINE__);
+	IRDMA_DI_REQ_HANDLER(peer, &req);
+}
+
 int
 irdma_register_qset(struct irdma_sc_vsi *vsi, struct irdma_ws_node *tc_node)
 {
@@ -611,32 +623,38 @@ irdma_dcqcn_tunables_init(struct irdma_pci_f *rf)
 		      &rf->dcqcn_params.min_rate, 0,
 		      "set minimum rate limit value, in MBits per second, default=0");
 
+	rf->dcqcn_params.dcqcn_f = 5;
 	SYSCTL_ADD_U8(&rf->tun_info.irdma_sysctl_ctx, irdma_sysctl_oid_list,
 		      OID_AUTO, "dcqcn_F", CTLFLAG_RDTUN, &rf->dcqcn_params.dcqcn_f, 0,
-		      "set number of times to stay in each stage of bandwidth recovery, default=0");
+		      "set number of times to stay in each stage of bandwidth recovery, default=5");
 
+	rf->dcqcn_params.dcqcn_t = 0x37;
 	SYSCTL_ADD_U16(&rf->tun_info.irdma_sysctl_ctx, irdma_sysctl_oid_list,
 		       OID_AUTO, "dcqcn_T", CTLFLAG_RDTUN, &rf->dcqcn_params.dcqcn_t, 0,
-		       "set number of usecs that should elapse before increasing the CWND in DCQCN mode, default=0");
+		       "set number of usecs that should elapse before increasing the CWND in DCQCN mode, default=0x37");
 
+	rf->dcqcn_params.dcqcn_b = 0x249f0;
 	SYSCTL_ADD_U32(&rf->tun_info.irdma_sysctl_ctx, irdma_sysctl_oid_list,
 		       OID_AUTO, "dcqcn_B", CTLFLAG_RDTUN, &rf->dcqcn_params.dcqcn_b, 0,
-		       "set number of MSS to add to the congestion window in additive increase mode, default=0");
+		       "set number of MSS to add to the congestion window in additive increase mode, default=0x249f0");
 
+	rf->dcqcn_params.rai_factor = 1;
 	SYSCTL_ADD_U16(&rf->tun_info.irdma_sysctl_ctx, irdma_sysctl_oid_list,
 		       OID_AUTO, "dcqcn_rai_factor", CTLFLAG_RDTUN,
 		       &rf->dcqcn_params.rai_factor, 0,
-		       "set number of MSS to add to the congestion window in additive increase mode, default=0");
+		       "set number of MSS to add to the congestion window in additive increase mode, default=1");
 
+	rf->dcqcn_params.hai_factor = 5;
 	SYSCTL_ADD_U16(&rf->tun_info.irdma_sysctl_ctx, irdma_sysctl_oid_list,
 		       OID_AUTO, "dcqcn_hai_factor", CTLFLAG_RDTUN,
 		       &rf->dcqcn_params.hai_factor, 0,
-		       "set number of MSS to add to the congestion window in hyperactive increase mode, default=0");
+		       "set number of MSS to add to the congestion window in hyperactive increase mode, default=5");
 
+	rf->dcqcn_params.rreduce_mperiod = 50;
 	SYSCTL_ADD_U32(&rf->tun_info.irdma_sysctl_ctx, irdma_sysctl_oid_list,
 		       OID_AUTO, "dcqcn_rreduce_mperiod", CTLFLAG_RDTUN,
 		       &rf->dcqcn_params.rreduce_mperiod, 0,
-		       "set minimum time between 2 consecutive rate reductions for a single flow, default=0");
+		       "set minimum time between 2 consecutive rate reductions for a single flow, default=50");
 }
 
 /**
@@ -743,3 +761,31 @@ irdma_prm_rem_bitmapmem(struct irdma_hw *hw, struct irdma_chunk *chunk)
 {
 	kfree(chunk->bitmapmem.va);
 }
+
+void
+irdma_cleanup_dead_qps(struct irdma_sc_vsi *vsi)
+{
+	struct irdma_sc_qp *qp = NULL;
+	struct irdma_qp *iwqp;
+	struct irdma_pci_f *rf;
+	u8 i;
+
+	for (i = 0; i < IRDMA_MAX_USER_PRIORITY; i++) {
+		qp = irdma_get_qp_from_list(&vsi->qos[i].qplist, qp);
+		while (qp) {
+			if (qp->qp_uk.qp_type == IRDMA_QP_TYPE_UDA) {
+				qp = irdma_get_qp_from_list(&vsi->qos[i].qplist, qp);
+				continue;
+			}
+			iwqp = qp->qp_uk.back_qp;
+			rf = iwqp->iwdev->rf;
+			irdma_free_dma_mem(rf->sc_dev.hw, &iwqp->q2_ctx_mem);
+			irdma_free_dma_mem(rf->sc_dev.hw, &iwqp->kqp.dma_mem);
+
+			kfree(iwqp->kqp.sq_wrid_mem);
+			kfree(iwqp->kqp.rq_wrid_mem);
+			qp = irdma_get_qp_from_list(&vsi->qos[i].qplist, qp);
+			kfree(iwqp);
+		}
+	}
+}
diff --git a/sys/dev/irdma/fbsd_kcompat.h b/sys/dev/irdma/fbsd_kcompat.h
@@ -40,10 +40,11 @@
 #define TASKLET_DATA_TYPE	unsigned long
 #define TASKLET_FUNC_TYPE	void (*)(TASKLET_DATA_TYPE)
 
+#ifndef tasklet_setup
 #define tasklet_setup(tasklet, callback)				\
 	tasklet_init((tasklet), (TASKLET_FUNC_TYPE)(callback),		\
 		      (TASKLET_DATA_TYPE)(tasklet))
-
+#endif
 #ifndef from_tasklet
 #define from_tasklet(var, callback_tasklet, tasklet_fieldname) \
 	container_of(callback_tasklet, typeof(*var), tasklet_fieldname)
@@ -176,8 +177,7 @@ int irdma_dereg_mr(struct ib_mr *ib_mr);
 #else
 int irdma_dereg_mr(struct ib_mr *ib_mr, struct ib_udata *udata);
 #endif
-void irdma_get_eth_speed_and_width(u32 link_speed, u8 *active_speed,
-				   u8 *active_width);
+int ib_get_eth_speed(struct ib_device *dev, u32 port_num, u8 *speed, u8 *width);
 enum rdma_link_layer irdma_get_link_layer(struct ib_device *ibdev,
 					  u8 port_num);
 int irdma_roce_port_immutable(struct ib_device *ibdev, u8 port_num,
@@ -197,6 +197,7 @@ int irdma_get_hw_stats(struct ib_device *ibdev,
 		       struct rdma_hw_stats *stats, u8 port_num,
 		       int index);
 
+void irdma_request_reset(struct irdma_pci_f *rf);
 int irdma_register_qset(struct irdma_sc_vsi *vsi,
 			struct irdma_ws_node *tc_node);
 void irdma_unregister_qset(struct irdma_sc_vsi *vsi,
@@ -337,4 +338,13 @@ static inline size_t irdma_ib_umem_num_dma_blocks(struct ib_umem *umem, unsigned
 			 ALIGN_DOWN(iova, pgsz))) / pgsz;
 }
 
+static inline void addrconf_addr_eui48(u8 *deui, const char *const addr)
+{
+	memcpy(deui, addr, 3);
+	deui[3] = 0xFF;
+	deui[4] = 0xFE;
+	memcpy(deui + 5, addr + 3, 3);
+	deui[0] ^= 2;
+}
+
 #endif /* FBSD_KCOMPAT_H */