IB/ehca: add Shared Receive Queue support

Support SRQs on eHCA2. Since an SRQ is a QP for eHCA2, a lot of code
(structures, create, destroy, post_recv) can be shared between QP and SRQ.
Signed-off-by: Joachim Fenkes <fenkes@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

drivers/infiniband/hw/ehca/ehca_classes.h

@@ -5,6 +5,7 @@
  *
  *  Authors: Heiko J Schick <schickhj@de.ibm.com>
  *           Christoph Raisch <raisch@de.ibm.com>
+ *           Joachim Fenkes <fenkes@de.ibm.com>
  *
  *  Copyright (c) 2005 IBM Corporation
  *
@@ -117,9 +118,20 @@ struct ehca_pd {
 	u32 ownpid;
 };
 
+enum ehca_ext_qp_type {
+	EQPT_NORMAL    = 0,
+	EQPT_LLQP      = 1,
+	EQPT_SRQBASE   = 2,
+	EQPT_SRQ       = 3,
+};
+
 struct ehca_qp {
+	union {
 		struct ib_qp ib_qp;
+		struct ib_srq ib_srq;
+	};
 	u32 qp_type;
+	enum ehca_ext_qp_type ext_type;
 	struct ipz_queue ipz_squeue;
 	struct ipz_queue ipz_rqueue;
 	struct h_galpas galpas;
@@ -142,6 +154,10 @@ struct ehca_qp {
 	u32 mm_count_galpa;
 };
 
+#define IS_SRQ(qp) (qp->ext_type == EQPT_SRQ)
+#define HAS_SQ(qp) (qp->ext_type != EQPT_SRQ)
+#define HAS_RQ(qp) (qp->ext_type != EQPT_SRQBASE)
+
 /* must be power of 2 */
 #define QP_HASHTAB_LEN 8
 
@@ -307,6 +323,7 @@ struct ehca_create_qp_resp {
 	u32 qp_num;
 	u32 token;
 	u32 qp_type;
+	u32 ext_type;
 	u32 qkey;
 	/* qp_num assigned by ehca: sqp0/1 may have got different numbers */
 	u32 real_qp_num;
@@ -329,13 +346,6 @@ enum ehca_service_type {
 	ST_UD  = 3,
 };
 
-enum ehca_ext_qp_type {
-	EQPT_NORMAL    = 0,
-	EQPT_LLQP      = 1,
-	EQPT_SRQBASE   = 2,
-	EQPT_SRQ       = 3,
-};
-
 enum ehca_ll_comp_flags {
 	LLQP_SEND_COMP = 0x20,
 	LLQP_RECV_COMP = 0x40,

drivers/infiniband/hw/ehca/ehca_classes_pSeries.h

@@ -228,8 +228,8 @@ struct hcp_modify_qp_control_block {
 #define MQPCB_QP_NUMBER                         EHCA_BMASK_IBM(8,31)
 #define MQPCB_MASK_QP_ENABLE                    EHCA_BMASK_IBM(48,48)
 #define MQPCB_QP_ENABLE                         EHCA_BMASK_IBM(31,31)
-#define MQPCB_MASK_CURR_SQR_LIMIT               EHCA_BMASK_IBM(49,49)
-#define MQPCB_CURR_SQR_LIMIT                    EHCA_BMASK_IBM(15,31)
+#define MQPCB_MASK_CURR_SRQ_LIMIT               EHCA_BMASK_IBM(49,49)
+#define MQPCB_CURR_SRQ_LIMIT                    EHCA_BMASK_IBM(16,31)
 #define MQPCB_MASK_QP_AFF_ASYN_EV_LOG_REG       EHCA_BMASK_IBM(50,50)
 #define MQPCB_MASK_SHARED_RQ_HNDL               EHCA_BMASK_IBM(51,51)
 

drivers/infiniband/hw/ehca/ehca_iverbs.h

@@ -154,6 +154,21 @@ int ehca_post_send(struct ib_qp *qp, struct ib_send_wr *send_wr,
 int ehca_post_recv(struct ib_qp *qp, struct ib_recv_wr *recv_wr,
 		   struct ib_recv_wr **bad_recv_wr);
 
+int ehca_post_srq_recv(struct ib_srq *srq,
+		       struct ib_recv_wr *recv_wr,
+		       struct ib_recv_wr **bad_recv_wr);
+
+struct ib_srq *ehca_create_srq(struct ib_pd *pd,
+			       struct ib_srq_init_attr *init_attr,
+			       struct ib_udata *udata);
+
+int ehca_modify_srq(struct ib_srq *srq, struct ib_srq_attr *attr,
+		    enum ib_srq_attr_mask attr_mask, struct ib_udata *udata);
+
+int ehca_query_srq(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
+
+int ehca_destroy_srq(struct ib_srq *srq);
+
 u64 ehca_define_sqp(struct ehca_shca *shca, struct ehca_qp *ibqp,
 		    struct ib_qp_init_attr *qp_init_attr);
 

drivers/infiniband/hw/ehca/ehca_main.c

@@ -343,7 +343,7 @@ int ehca_init_device(struct ehca_shca *shca)
 	strlcpy(shca->ib_device.name, "ehca%d", IB_DEVICE_NAME_MAX);
 	shca->ib_device.owner               = THIS_MODULE;
 
-	shca->ib_device.uverbs_abi_ver	    = 6;
+	shca->ib_device.uverbs_abi_ver	    = 7;
 	shca->ib_device.uverbs_cmd_mask	    =
 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
@@ -411,6 +411,20 @@ int ehca_init_device(struct ehca_shca *shca)
 	/* shca->ib_device.process_mad	    = ehca_process_mad;	    */
 	shca->ib_device.mmap		    = ehca_mmap;
 
+	if (EHCA_BMASK_GET(HCA_CAP_SRQ, shca->hca_cap)) {
+		shca->ib_device.uverbs_cmd_mask |=
+			(1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
+			(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
+			(1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
+			(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ);
+
+		shca->ib_device.create_srq          = ehca_create_srq;
+		shca->ib_device.modify_srq          = ehca_modify_srq;
+		shca->ib_device.query_srq           = ehca_query_srq;
+		shca->ib_device.destroy_srq         = ehca_destroy_srq;
+		shca->ib_device.post_srq_recv       = ehca_post_srq_recv;
+	}
+
 	return ret;
 }
 

drivers/infiniband/hw/ehca/ehca_reqs.c

@@ -3,8 +3,9 @@
  *
  *  post_send/recv, poll_cq, req_notify
  *
- *  Authors: Waleri Fomin <fomin@de.ibm.com>
- *           Hoang-Nam Nguyen <hnguyen@de.ibm.com>
+ *  Authors: Hoang-Nam Nguyen <hnguyen@de.ibm.com>
+ *           Waleri Fomin <fomin@de.ibm.com>
+ *           Joachim Fenkes <fenkes@de.ibm.com>
  *           Reinhard Ernst <rernst@de.ibm.com>
  *
  *  Copyright (c) 2005 IBM Corporation
@@ -413,17 +414,23 @@ int ehca_post_send(struct ib_qp *qp,
 	return ret;
 }
 
-int ehca_post_recv(struct ib_qp *qp,
+static int internal_post_recv(struct ehca_qp *my_qp,
+			      struct ib_device *dev,
 			      struct ib_recv_wr *recv_wr,
 			      struct ib_recv_wr **bad_recv_wr)
 {
-	struct ehca_qp *my_qp = container_of(qp, struct ehca_qp, ib_qp);
 	struct ib_recv_wr *cur_recv_wr;
 	struct ehca_wqe *wqe_p;
 	int wqe_cnt = 0;
 	int ret = 0;
 	unsigned long spl_flags;
 
+	if (unlikely(!HAS_RQ(my_qp))) {
+		ehca_err(dev, "QP has no RQ  ehca_qp=%p qp_num=%x ext_type=%d",
+			 my_qp, my_qp->real_qp_num, my_qp->ext_type);
+		return -ENODEV;
+	}
+
 	/* LOCK the QUEUE */
 	spin_lock_irqsave(&my_qp->spinlock_r, spl_flags);
 
@@ -439,8 +446,8 @@ int ehca_post_recv(struct ib_qp *qp,
 				*bad_recv_wr = cur_recv_wr;
 			if (wqe_cnt == 0) {
 				ret = -ENOMEM;
-				ehca_err(qp->device, "Too many posted WQEs "
-					 "qp_num=%x", qp->qp_num);
+				ehca_err(dev, "Too many posted WQEs "
+					 "qp_num=%x", my_qp->real_qp_num);
 			}
 			goto post_recv_exit0;
 		}
@@ -455,14 +462,14 @@ int ehca_post_recv(struct ib_qp *qp,
 			*bad_recv_wr = cur_recv_wr;
 			if (wqe_cnt == 0) {
 				ret = -EINVAL;
-				ehca_err(qp->device, "Could not write WQE "
-					 "qp_num=%x", qp->qp_num);
+				ehca_err(dev, "Could not write WQE "
+					 "qp_num=%x", my_qp->real_qp_num);
 			}
 			goto post_recv_exit0;
 		}
 		wqe_cnt++;
-		ehca_gen_dbg("ehca_qp=%p qp_num=%x wqe_cnt=%d",
-		     my_qp, qp->qp_num, wqe_cnt);
+		ehca_dbg(dev, "ehca_qp=%p qp_num=%x wqe_cnt=%d",
+			 my_qp, my_qp->real_qp_num, wqe_cnt);
 	} /* eof for cur_recv_wr */
 
 post_recv_exit0:
@@ -472,6 +479,22 @@ int ehca_post_recv(struct ib_qp *qp,
 	return ret;
 }
 
+int ehca_post_recv(struct ib_qp *qp,
+		   struct ib_recv_wr *recv_wr,
+		   struct ib_recv_wr **bad_recv_wr)
+{
+	return internal_post_recv(container_of(qp, struct ehca_qp, ib_qp),
+				  qp->device, recv_wr, bad_recv_wr);
+}
+
+int ehca_post_srq_recv(struct ib_srq *srq,
+		       struct ib_recv_wr *recv_wr,
+		       struct ib_recv_wr **bad_recv_wr)
+{
+	return internal_post_recv(container_of(srq, struct ehca_qp, ib_srq),
+				  srq->device, recv_wr, bad_recv_wr);
+}
+
 /*
  * ib_wc_opcode table converts ehca wc opcode to ib
  * Since we use zero to indicate invalid opcode, the actual ib opcode must

drivers/infiniband/hw/ehca/ehca_uverbs.c

@@ -257,6 +257,7 @@ int ehca_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 	struct ehca_cq *cq;
 	struct ehca_qp *qp;
 	struct ehca_pd *pd;
+	struct ib_uobject *uobject;
 
 	switch (q_type) {
 	case  1: /* CQ */
@@ -304,7 +305,8 @@ int ehca_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 			return -ENOMEM;
 		}
 
-		if (!qp->ib_qp.uobject || qp->ib_qp.uobject->context != context)
+		uobject = IS_SRQ(qp) ? qp->ib_srq.uobject : qp->ib_qp.uobject;
+		if (!uobject || uobject->context != context)
 			return -EINVAL;
 
 		ret = ehca_mmap_qp(vma, qp, rsrc_type);

drivers/infiniband/hw/ehca/hcp_if.c

@@ -5,6 +5,7 @@
  *
  *  Authors: Christoph Raisch <raisch@de.ibm.com>
  *           Hoang-Nam Nguyen <hnguyen@de.ibm.com>
+ *           Joachim Fenkes <fenkes@de.ibm.com>
  *           Gerd Bayer <gerd.bayer@de.ibm.com>
  *           Waleri Fomin <fomin@de.ibm.com>
  *
@@ -62,6 +63,12 @@
 #define H_ALL_RES_QP_MAX_SEND_SGE       EHCA_BMASK_IBM(32, 39)
 #define H_ALL_RES_QP_MAX_RECV_SGE       EHCA_BMASK_IBM(40, 47)
 
+#define H_ALL_RES_QP_UD_AV_LKEY         EHCA_BMASK_IBM(32, 63)
+#define H_ALL_RES_QP_SRQ_QP_TOKEN       EHCA_BMASK_IBM(0, 31)
+#define H_ALL_RES_QP_SRQ_QP_HANDLE      EHCA_BMASK_IBM(0, 64)
+#define H_ALL_RES_QP_SRQ_LIMIT          EHCA_BMASK_IBM(48, 63)
+#define H_ALL_RES_QP_SRQ_QPN            EHCA_BMASK_IBM(40, 63)
+
 #define H_ALL_RES_QP_ACT_OUTST_SEND_WR  EHCA_BMASK_IBM(16, 31)
 #define H_ALL_RES_QP_ACT_OUTST_RECV_WR  EHCA_BMASK_IBM(48, 63)
 #define H_ALL_RES_QP_ACT_SEND_SGE       EHCA_BMASK_IBM(8, 15)
@@ -150,7 +157,7 @@ static long ehca_plpar_hcall9(unsigned long opcode,
 {
 	long ret;
 	int i, sleep_msecs, lock_is_set = 0;
-	unsigned long flags;
+	unsigned long flags = 0;
 
 	ehca_gen_dbg("opcode=%lx arg1=%lx arg2=%lx arg3=%lx arg4=%lx "
 		     "arg5=%lx arg6=%lx arg7=%lx arg8=%lx arg9=%lx",
@@ -282,8 +289,7 @@ u64 hipz_h_alloc_resource_qp(const struct ipz_adapter_handle adapter_handle,
 			     struct ehca_alloc_qp_parms *parms)
 {
 	u64 ret;
-	u64 allocate_controls;
-	u64 max_r10_reg;
+	u64 allocate_controls, max_r10_reg, r11, r12;
 	u64 outs[PLPAR_HCALL9_BUFSIZE];
 
 	allocate_controls =
@@ -309,6 +315,13 @@ u64 hipz_h_alloc_resource_qp(const struct ipz_adapter_handle adapter_handle,
 		| EHCA_BMASK_SET(H_ALL_RES_QP_MAX_RECV_SGE,
 				 parms->max_recv_sge);
 
+	r11 = EHCA_BMASK_SET(H_ALL_RES_QP_SRQ_QP_TOKEN, parms->srq_token);
+
+	if (parms->ext_type == EQPT_SRQ)
+		r12 = EHCA_BMASK_SET(H_ALL_RES_QP_SRQ_LIMIT, parms->srq_limit);
+	else
+		r12 = EHCA_BMASK_SET(H_ALL_RES_QP_SRQ_QPN, parms->srq_qpn);
+
 	ret = ehca_plpar_hcall9(H_ALLOC_RESOURCE, outs,
 				adapter_handle.handle,	           /* r4  */
 				allocate_controls,	           /* r5  */
@@ -316,9 +329,7 @@ u64 hipz_h_alloc_resource_qp(const struct ipz_adapter_handle adapter_handle,
 				parms->recv_cq_handle.handle,
 				parms->eq_handle.handle,
 				((u64)parms->token << 32) | parms->pd.value,
-				max_r10_reg,	                   /* r10 */
-				parms->ud_av_l_key_ctl,            /* r11 */
-				0);
+				max_r10_reg, r11, r12);
 
 	parms->qp_handle.handle = outs[0];
 	parms->real_qp_num = (u32)outs[1];

drivers/infiniband/hw/ehca/hipz_hw.h

@@ -163,6 +163,7 @@ struct hipz_qptemm {
 
 #define QPX_SQADDER EHCA_BMASK_IBM(48,63)
 #define QPX_RQADDER EHCA_BMASK_IBM(48,63)
+#define QPX_AAELOG_RESET_SRQ_LIMIT EHCA_BMASK_IBM(3,3)
 
 #define QPTEMM_OFFSET(x) offsetof(struct hipz_qptemm,x)