Misc cleanups and implement rereg_user_mr
Subsystems (2)
cover Misc cleanups and implement rereg_user_mr
This patch set does some preparatory cleanups and then implements the two simple cases of ib_rereg_user_mr.
Bob Pearson (6):
RDMA/rxe: Rename IB_ACCESS_REMOTE rdma/rxe: Optimize send path in rxe_resp.c RDMA/RXE: Fix access checks in rxe_check_bind_mw RDMA/RXE: Introduce rxe access supported flags RDMA/RXE: Let rkey == lkey for local access RDMA/RXE: Implement rereg_user_mr
drivers/infiniband/sw/rxe/rxe_mr.c | 21 ++++++------- drivers/infiniband/sw/rxe/rxe_mw.c | 22 +++++++++----- drivers/infiniband/sw/rxe/rxe_opcode.h | 3 ++ drivers/infiniband/sw/rxe/rxe_qp.c | 7 +++++ drivers/infiniband/sw/rxe/rxe_resp.c | 12 ++++++-- drivers/infiniband/sw/rxe/rxe_verbs.c | 41 ++++++++++++++++++++++++++ drivers/infiniband/sw/rxe/rxe_verbs.h | 20 +++++++++++++ 7 files changed, 104 insertions(+), 22 deletions(-)
base-commit: 8c1ee346da583718fb0a7791a1f84bdafb103caf
Signature · 1 line
2.39.2
PATCH 1/6 RDMA/rxe: Rename IB_ACCESS_REMOTE
Rename IB_ACCESS_REMOTE to RXE_ACCESS_REMOTE and move to rxe_verbs.h as an enum instead of a #define. Shouldn't use IB_xxx for rxe symbols.
--- drivers/infiniband/sw/rxe/rxe_mr.c | 10 +++------- drivers/infiniband/sw/rxe/rxe_verbs.h | 6 ++++++ 2 files changed, 9 insertions(+), 7 deletions(-)
drivers/infiniband/sw/rxe/rxe_mr.c +3 -7
@@ -45,14 +45,10 @@ int mr_check_range(struct rxe_mr *mr, u64 iova, size_t length) } } -#define IB_ACCESS_REMOTE (IB_ACCESS_REMOTE_READ \ - | IB_ACCESS_REMOTE_WRITE \ - | IB_ACCESS_REMOTE_ATOMIC) -static void rxe_mr_init (int access,struct rxe_mr *mr) {u32 lkey = mr->elem.index <<8 |rxe_get_next_key (-1 ); -u32 rkey = (access &IB_ACCESS_REMOTE ) ? lkey :0 ; +u32 rkey = (access &RXE_ACCESS_REMOTE ) ? lkey :0 ;/* set ibmr->l/rkey and also copy into private l/rkey * for user MRs these will always be the same
@@ -195,7 +191,7 @@ int rxe_mr_init_fast(int max_pages, struct rxe_mr *mr)int err;/* always allow remote access for FMRs */ -rxe_mr_init (IB_ACCESS_REMOTE , mr); +rxe_mr_init (RXE_ACCESS_REMOTE , mr); err =rxe_mr_alloc (mr, max_pages);if (err)
@@ -715,7 +711,7 @@ int rxe_reg_fast_mr(struct rxe_qp *qp, struct rxe_send_wqe *wqe) mr->access = access; mr->lkey = key; - mr->rkey = (access &IB_ACCESS_REMOTE ) ? key :0 ; + mr->rkey = (access &RXE_ACCESS_REMOTE ) ? key :0 ; mr->ibmr.iova = wqe->wr.wr.reg.mr->iova; mr->state =RXE_MR_STATE_VALID ;
drivers/infiniband/sw/rxe/rxe_verbs.h +6 -0
@@ -253,6 +253,12 @@ struct rxe_qp {struct execute_work cleanup_work; }; +enum rxe_access { +RXE_ACCESS_REMOTE = (IB_ACCESS_REMOTE_READ + |IB_ACCESS_REMOTE_WRITE + |IB_ACCESS_REMOTE_ATOMIC ), +}; +enum rxe_mr_state {RXE_MR_STATE_INVALID ,RXE_MR_STATE_FREE ,
Signature · 1 line
2.39.2
PATCH 2/6 rdma/rxe: Optimize send path in rxe_resp.c
In reply to Bob PearsonBypass calling check_rkey() in rxe_resp.c for non-rdma messages.
--- drivers/infiniband/sw/rxe/rxe_opcode.h | 3 +++ drivers/infiniband/sw/rxe/rxe_resp.c | 12 ++++++++++-- 2 files changed, 13 insertions(+), 2 deletions(-)
drivers/infiniband/sw/rxe/rxe_opcode.h +3 -0
@@ -91,6 +91,9 @@ enum rxe_hdr_mask {RXE_READ_OR_ATOMIC_MASK = (RXE_READ_MASK |RXE_ATOMIC_MASK ),RXE_WRITE_OR_SEND_MASK = (RXE_WRITE_MASK |RXE_SEND_MASK ),RXE_READ_OR_WRITE_MASK = (RXE_READ_MASK |RXE_WRITE_MASK ), +RXE_RDMA_OP_MASK = (RXE_READ_MASK |RXE_WRITE_MASK | +RXE_ATOMIC_WRITE_MASK |RXE_FLUSH_MASK | + RXE_ATOMIC_MASK), };#define OPCODE_NONE (-1)
drivers/infiniband/sw/rxe/rxe_resp.c +10 -2
@@ -387,7 +387,10 @@ static enum resp_states rxe_resp_check_length(struct rxe_qp *qp, } } -return RESPST_CHK_RKEY ; +if (pkt->mask &RXE_RDMA_OP_MASK ) +return RESPST_CHK_RKEY ; +else +return RESPST_EXECUTE ; }/* if the reth length field is zero we can assume nothing
@@ -434,6 +437,10 @@ static enum resp_states check_rkey(struct rxe_qp *qp,enum resp_states state;int access =0 ; +/* parse RETH or ATMETH header for first/only packets +* for va, length, rkey, etc. or use current value for +* middle/last packets. +*/ if (pkt->mask & (RXE_READ_OR_WRITE_MASK |RXE_ATOMIC_WRITE_MASK )) {if (pkt->mask &RXE_RETH_MASK )qp_resp_from_reth (qp, pkt);
@@ -454,7 +461,8 @@ static enum resp_states check_rkey(struct rxe_qp *qp,qp_resp_from_atmeth (qp, pkt); access =IB_ACCESS_REMOTE_ATOMIC ; }else { -return RESPST_EXECUTE ; +/* shouldn't happen */ +WARN_ON (1 ); }/* A zero-byte read or write op is not required to
Signature · 1 line
2.39.2
PATCH 3/6 RDMA/rxe: Fix access checks in rxe_check_bind_mw
In reply to Bob PearsonThe subroutine rxe_check_bind_mw() in rxe_mw.c performs checks on the mw access flags before they are set so they always succeed. This patch instead checks the access flags passed in the send wqe.
--- drivers/infiniband/sw/rxe/rxe_mw.c | 17 +++++++++-------- 1 file changed, 9 insertions(+), 8 deletions(-)
drivers/infiniband/sw/rxe/rxe_mw.c +9 -8
@@ -48,7 +48,7 @@ int rxe_dealloc_mw(struct ib_mw *ibmw) }static int rxe_check_bind_mw (struct rxe_qp *qp,struct rxe_send_wqe *wqe, -struct rxe_mw *mw,struct rxe_mr *mr) +struct rxe_mw *mw,struct rxe_mr *mr,int access) {if (mw->ibmw.type ==IB_MW_TYPE_1 ) {if (unlikely (mw->state !=RXE_MW_STATE_VALID )) {
@@ -58,7 +58,7 @@ static int rxe_check_bind_mw(struct rxe_qp *qp, struct rxe_send_wqe *wqe, }/* o10-36.2.2 */ -if (unlikely ((mw->access &IB_ZERO_BASED ))) { +if (unlikely ((access &IB_ZERO_BASED ))) {rxe_dbg_mw (mw,"attempt to bind a zero based type 1 MW\n" );return -EINVAL ; }
@@ -104,7 +104,7 @@ static int rxe_check_bind_mw(struct rxe_qp *qp, struct rxe_send_wqe *wqe, }/* C10-74 */ -if (unlikely ((mw->access & +if (unlikely ((access & (IB_ACCESS_REMOTE_WRITE |IB_ACCESS_REMOTE_ATOMIC )) && !(mr->access &IB_ACCESS_LOCAL_WRITE ))) { rxe_dbg_mw(mw,
@@ -113,7 +113,7 @@ static int rxe_check_bind_mw(struct rxe_qp *qp, struct rxe_send_wqe *wqe, }/* C10-75 */ -if (mw->access &IB_ZERO_BASED ) { +if (access &IB_ZERO_BASED ) {if (unlikely (wqe->wr.wr.mw.length > mr->ibmr.length)) {rxe_dbg_mw (mw,"attempt to bind a ZB MW outside of the MR\n" );
@@ -133,12 +133,12 @@ static int rxe_check_bind_mw(struct rxe_qp *qp, struct rxe_send_wqe *wqe, }static void rxe_do_bind_mw (struct rxe_qp *qp,struct rxe_send_wqe *wqe, -struct rxe_mw *mw,struct rxe_mr *mr) +struct rxe_mw *mw,struct rxe_mr *mr,int access) {u32 key = wqe->wr.wr.mw.rkey &0xff ; mw->rkey = (mw->rkey & ~0xff ) | key; - mw->access = wqe->wr.wr.mw.access; + mw->access = access; mw->state =RXE_MW_STATE_VALID ; mw->addr = wqe->wr.wr.mw.addr; mw->length = wqe->wr.wr.mw.length;
@@ -169,6 +169,7 @@ int rxe_bind_mw(struct rxe_qp *qp, struct rxe_send_wqe *wqe)struct rxe_dev *rxe =to_rdev (qp->ibqp.device);u32 mw_rkey = wqe->wr.wr.mw.mw_rkey;u32 mr_lkey = wqe->wr.wr.mw.mr_lkey; +int access = wqe->wr.wr.mw.access; mw =rxe_pool_get_index (&rxe->mw_pool, mw_rkey >>8 );if (unlikely (!mw)) {
@@ -198,11 +199,11 @@ int rxe_bind_mw(struct rxe_qp *qp, struct rxe_send_wqe *wqe)spin_lock_bh (&mw->lock); - ret =rxe_check_bind_mw (qp, wqe, mw, mr); + ret =rxe_check_bind_mw (qp, wqe, mw, mr, access);if (ret) goto err_unlock; -rxe_do_bind_mw (qp, wqe, mw, mr); +rxe_do_bind_mw (qp, wqe, mw, mr, access); err_unlock:spin_unlock_bh (&mw->lock); err_drop_mr:
Signature · 1 line
2.39.2
PATCH 4/6 RDMA/rxe: Introduce rxe access supported flags
In reply to Bob PearsonIntroduce supported bit masks for setting the access attributes of MWs, MRs, and QPs. Check these when attributes are set.
--- drivers/infiniband/sw/rxe/rxe_mw.c | 5 +++++ drivers/infiniband/sw/rxe/rxe_qp.c | 7 +++++++ drivers/infiniband/sw/rxe/rxe_verbs.c | 6 ++++++ drivers/infiniband/sw/rxe/rxe_verbs.h | 15 ++++++++++++--- 4 files changed, 30 insertions(+), 3 deletions(-)
drivers/infiniband/sw/rxe/rxe_mw.c +5 -0
@@ -197,6 +197,11 @@ int rxe_bind_mw(struct rxe_qp *qp, struct rxe_send_wqe *wqe) mr =NULL ; } +if (access & ~RXE_ACCESS_SUPPORTED_MW ) { +rxe_err_mw (mw,"access %#x not supported" , access); +return -EOPNOTSUPP ; + } +spin_lock_bh (&mw->lock); ret =rxe_check_bind_mw (qp, wqe, mw, mr, access);
drivers/infiniband/sw/rxe/rxe_qp.c +7 -0
@@ -392,6 +392,13 @@ int rxe_qp_chk_attr(struct rxe_dev *rxe, struct rxe_qp *qp,if (mask &IB_QP_CAP &&rxe_qp_chk_cap (rxe, &attr->cap, !!qp->srq)) goto err1; +if (mask &IB_QP_ACCESS_FLAGS ) { +if (!(qp_type (qp) ==IB_QPT_RC ||qp_type (qp) ==IB_QPT_UC )) + goto err1; +if (attr->qp_access_flags & ~RXE_ACCESS_SUPPORTED_QP ) + goto err1; + } +if (mask &IB_QP_AV &&rxe_av_chk_attr (qp, &attr->ah_attr)) goto err1;
drivers/infiniband/sw/rxe/rxe_verbs.c +6 -0
@@ -1260,6 +1260,12 @@ static struct ib_mr *rxe_reg_user_mr(struct ib_pd *ibpd, u64 start,struct rxe_mr *mr;int err, cleanup_err; +if (access & ~RXE_ACCESS_SUPPORTED_MR ) { +rxe_err_pd (pd,"access = %#x not supported (%#x)" , access, +RXE_ACCESS_SUPPORTED_MR ); +return ERR_PTR (-EOPNOTSUPP ); + } + mr =kzalloc (sizeof (*mr),GFP_KERNEL );if (!mr)return ERR_PTR (-ENOMEM );
drivers/infiniband/sw/rxe/rxe_verbs.h +12 -3
@@ -253,10 +253,19 @@ struct rxe_qp {struct execute_work cleanup_work; }; -enum rxe_access { -RXE_ACCESS_REMOTE = (IB_ACCESS_REMOTE_READ +enum { +RXE_ACCESS_REMOTE =IB_ACCESS_REMOTE_READ |IB_ACCESS_REMOTE_WRITE - |IB_ACCESS_REMOTE_ATOMIC ), + |IB_ACCESS_REMOTE_ATOMIC , +RXE_ACCESS_SUPPORTED_MR =RXE_ACCESS_REMOTE + |IB_ACCESS_LOCAL_WRITE + |IB_ACCESS_MW_BIND + |IB_ACCESS_ON_DEMAND + |IB_ACCESS_FLUSH_GLOBAL + |IB_ACCESS_FLUSH_PERSISTENT , +RXE_ACCESS_SUPPORTED_QP =RXE_ACCESS_SUPPORTED_MR , +RXE_ACCESS_SUPPORTED_MW =RXE_ACCESS_SUPPORTED_MR + |IB_ZERO_BASED , };enum rxe_mr_state {
Signature · 1 line
2.39.2
PATCH 5/6 RDMA/rxe: Let rkey == lkey for local access
In reply to Bob PearsonIn order to conform to other drivers stop using rkey == 0 as an indication that there are no remote access flags set. Set rkey == lkey by default for all MRs.
--- drivers/infiniband/sw/rxe/rxe_mr.c | 15 ++++++++------- 1 file changed, 8 insertions(+), 7 deletions(-)
drivers/infiniband/sw/rxe/rxe_mr.c +8 -7
@@ -47,16 +47,15 @@ int mr_check_range(struct rxe_mr *mr, u64 iova, size_t length)static void rxe_mr_init (int access,struct rxe_mr *mr) { -u32 lkey = mr->elem.index <<8 |rxe_get_next_key (-1 ); -u32 rkey = (access &RXE_ACCESS_REMOTE ) ? lkey :0 ; +u32 key = mr->elem.index <<8 |rxe_get_next_key (-1 );/* set ibmr->l/rkey and also copy into private l/rkey * for user MRs these will always be the same * for cases where caller 'owns' the key portion * they may be different until REG_MR WQE is executed. */ - mr->lkey = mr->ibmr.lkey = lkey; - mr->rkey = mr->ibmr.rkey = rkey; + mr->lkey = mr->ibmr.lkey = key; + mr->rkey = mr->ibmr.rkey = key; mr->access = access; mr->ibmr.page_size =PAGE_SIZE ;
@@ -640,6 +639,7 @@ int rxe_invalidate_mr(struct rxe_qp *qp, u32 key) {struct rxe_dev *rxe =to_rdev (qp->ibqp.device);struct rxe_mr *mr; +int remote;int ret; mr =rxe_pool_get_index (&rxe->mr_pool, key >>8 );
@@ -649,9 +649,10 @@ int rxe_invalidate_mr(struct rxe_qp *qp, u32 key) goto err; } -if (mr->rkey ? (key != mr->rkey) : (key != mr->lkey)) { + remote = mr->access &RXE_ACCESS_REMOTE ; +if (remote ? (key != mr->rkey) : (key != mr->lkey)) {rxe_dbg_mr (mr,"wr key (%#x) doesn't match mr key (%#x)\n" , - key, (mr->rkey ? mr->rkey : mr->lkey)); + key, (remote ? mr->rkey : mr->lkey)); ret = -EINVAL ; goto err_drop_ref; }
@@ -711,7 +712,7 @@ int rxe_reg_fast_mr(struct rxe_qp *qp, struct rxe_send_wqe *wqe) mr->access = access; mr->lkey = key; - mr->rkey = (access &RXE_ACCESS_REMOTE ) ? key :0 ; + mr->rkey = key; mr->ibmr.iova = wqe->wr.wr.reg.mr->iova; mr->state =RXE_MR_STATE_VALID ;
Signature · 1 line
2.39.2
PATCH 6/6 RDMA/rxe: Implement rereg_user_mr
In reply to Bob PearsonImplement the two easy cases of ib_rereg_user_mr.
--- drivers/infiniband/sw/rxe/rxe_verbs.c | 35 +++++++++++++++++++++++++++ drivers/infiniband/sw/rxe/rxe_verbs.h | 5 ++++ 2 files changed, 40 insertions(+)
drivers/infiniband/sw/rxe/rxe_verbs.c +35 -0
@@ -1299,6 +1299,40 @@ static struct ib_mr *rxe_reg_user_mr(struct ib_pd *ibpd, u64 start,return ERR_PTR (err); } +static struct ib_mr *rxe_rereg_user_mr (struct ib_mr *ibmr,int flags, +u64 start,u64 length,u64 iova, +int access,struct ib_pd *ibpd, +struct ib_udata *udata) +{ +struct rxe_mr *mr =to_rmr (ibmr); +struct rxe_pd *old_pd =to_rpd (ibmr->pd); +struct rxe_pd *pd =to_rpd (ibpd); + +/* for now only support the two easy cases: +* rereg_pd and rereg_access +*/ +if (flags & ~RXE_MR_REREG_SUPPORTED ) { +rxe_err_mr (mr,"flags = %#x not supported" , flags); +return ERR_PTR (-EOPNOTSUPP ); + } + +if (flags &IB_MR_REREG_PD ) { +rxe_put (old_pd); +rxe_get (pd); + mr->ibmr.pd = ibpd; + } + +if (flags &IB_MR_REREG_ACCESS ) { +if (access & ~RXE_ACCESS_SUPPORTED_MR ) { +rxe_err_mr (mr,"access = %#x not supported" , access); +return ERR_PTR (-EOPNOTSUPP ); + } + mr->access = access; + } + +return NULL ; +} +static struct ib_mr *rxe_alloc_mr (struct ib_pd *ibpd,enum ib_mr_type mr_type,u32 max_num_sg) {
@@ -1451,6 +1485,7 @@ static const struct ib_device_ops rxe_dev_ops = { .query_srq = rxe_query_srq, .reg_user_mr = rxe_reg_user_mr, .req_notify_cq = rxe_req_notify_cq, + .rereg_user_mr = rxe_rereg_user_mr, .resize_cq = rxe_resize_cq, INIT_RDMA_OBJ_SIZE(ib_ah, rxe_ah, ibah),
drivers/infiniband/sw/rxe/rxe_verbs.h +5 -0
@@ -284,6 +284,11 @@ enum rxe_mr_lookup_type { RXE_LOOKUP_REMOTE, }; +enum rxe_rereg { +RXE_MR_REREG_SUPPORTED =IB_MR_REREG_PD + |IB_MR_REREG_ACCESS , +}; +static inline int rkey_is_mw (u32 rkey) {u32 index = rkey >>8 ;
Signature · 1 line
2.39.2
Quoted 10 lines · Bob Pearson
This patch set does some preparatory cleanups and then implements the two simple cases of
ib_rereg_user_mr.Bob Pearson (6):
RDMA/rxe: RenameIB_ACCESS_REMOTErdma/rxe: Optimize send path inrxe_resp.cRDMA/RXE: Fix access checks inrxe_check_bind_mwRDMA/RXE: Introduce rxe access supported flags RDMA/RXE: Let rkey == lkey for local access RDMA/RXE: Implementrereg_user_mr
Applied to for-next, thanks
Jason