RDMA/rxe: Split rxe_invalidate_mr into local and remote versions
Subsystems (2)
PATCH v2 1/1 RDMA/rxe: Split rxe_invalidate_mr into local and remote versions
Currently rxe_invalidate_mr does invalidate for both local ops, and remote ones. This means that MR being invalidated is compared with rkey for both, which is incorrect. For local invalidate, comparison should happen with lkey, and for the remote one, it should happen with rkey.
This commit splits the rxe_invalidate_mr into local and remote versions. Each of them does comparison the right way as described above (with lkey for local, and rkey for remote).
---
v1 -> v2
give a better name to key variable in function rxe_do_local_ops
drivers/infiniband/sw/rxe/rxe_loc.h | 3 +- drivers/infiniband/sw/rxe/rxe_mr.c | 59 +++++++++++++++++++++------- drivers/infiniband/sw/rxe/rxe_req.c | 10 ++--- drivers/infiniband/sw/rxe/rxe_resp.c | 2 +- 4 files changed, 53 insertions(+), 21 deletions(-)
drivers/infiniband/sw/rxe/rxe_loc.h +2 -1
@@ -77,7 +77,8 @@ struct rxe_mr *lookup_mr(struct rxe_pd *pd, int access, u32 key, enum rxe_mr_lookup_type type);int mr_check_range (struct rxe_mr *mr,u64 iova,size_t length);int advance_dma_data (struct rxe_dma_info *dma,unsigned int length); -int rxe_invalidate_mr (struct rxe_qp *qp,u32 rkey); +int rxe_invalidate_mr_local (struct rxe_qp *qp,u32 lkey); +int rxe_invalidate_mr_remote (struct rxe_qp *qp,u32 rkey);int rxe_reg_fast_mr (struct rxe_qp *qp,struct rxe_send_wqe *wqe);int rxe_mr_set_page (struct ib_mr *ibmr,u64 addr);int rxe_dereg_mr (struct ib_mr *ibmr,struct ib_udata *udata);
drivers/infiniband/sw/rxe/rxe_mr.c +45 -14
86 lines not on this page · raw message
drivers/infiniband/sw/rxe/rxe_req.c +5 -5
@@ -541,16 +541,16 @@ static void update_state(struct rxe_qp *qp, struct rxe_pkt_info *pkt)static int rxe_do_local_ops (struct rxe_qp *qp,struct rxe_send_wqe *wqe) {u8 opcode = wqe->wr.opcode; -u32 rkey; +u32 key;int ret;switch (opcode) {case IB_WR_LOCAL_INV : - rkey = wqe->wr.ex.invalidate_rkey; -if (rkey_is_mw (rkey)) - ret =rxe_invalidate_mw (qp, rkey); + key = wqe->wr.ex.invalidate_rkey; +if (rkey_is_mw (key)) + ret =rxe_invalidate_mw (qp, key);else - ret =rxe_invalidate_mr (qp, rkey); + ret =rxe_invalidate_mr_local (qp, key);if (unlikely (ret)) { wqe->status =IB_WC_LOC_QP_OP_ERR ;
drivers/infiniband/sw/rxe/rxe_resp.c +1 -1
@@ -818,7 +818,7 @@ static int invalidate_rkey(struct rxe_qp *qp, u32 rkey)if (rkey_is_mw (rkey))return rxe_invalidate_mw (qp, rkey);else -return rxe_invalidate_mr (qp, rkey); +return rxe_invalidate_mr_remote (qp, rkey); }/* Executes a new request. A retried request never reach that function (send
Signature · 1 line
2.25.1
Quoted 168 lines · Md Haris Iqbal
Currently
rxe_invalidate_mrdoes invalidate for both local ops, and remote ones. This means that MR being invalidated is compared with rkey for both, which is incorrect. For local invalidate, comparison should happen with lkey, and for the remote one, it should happen with rkey.This commit splits the
rxe_invalidate_mrinto local and remote versions. Each of them does comparison the right way as described above (with lkey for local, and rkey for remote).Fixes 3902b429ca14 ("RDMA/rxe: Implement invalidate MW operations")Signed-off-by Md Haris Iqbal <[email protected]>---
v1 -> v2
give a better name to key variable in functionrxe_do_local_opsdrivers/infiniband/sw/rxe/rxe_loc.h | 3 +- drivers/infiniband/sw/rxe/rxe_mr.c | 59 +++++++++++++++++++++------- drivers/infiniband/sw/rxe/rxe_req.c | 10 ++--- drivers/infiniband/sw/rxe/rxe_resp.c | 2 +- 4 files changed, 53 insertions(+), 21 deletions(-)drivers/infiniband/sw/rxe/rxe_loc.h +2
-1@@ -77,7 +77,8 @@ struct rxe_mr *lookup_mr(struct rxe_pd *pd, int access, u32 key, enum rxe_mr_lookup_type type);int mr_check_range (struct rxe_mr *mr,u64 iova,size_t length);int advance_dma_data (struct rxe_dma_info *dma,unsigned int length); -int rxe_invalidate_mr (struct rxe_qp *qp,u32 rkey); +int rxe_invalidate_mr_local (struct rxe_qp *qp,u32 lkey); +int rxe_invalidate_mr_remote (struct rxe_qp *qp,u32 rkey);int rxe_reg_fast_mr (struct rxe_qp *qp,struct rxe_send_wqe *wqe);int rxe_mr_set_page (struct ib_mr *ibmr,u64 addr);int rxe_dereg_mr (struct ib_mr *ibmr,struct ib_udata *udata);drivers/infiniband/sw/rxe/rxe_mr.c +45
-14@@ -576,41 +576,72 @@ struct rxe_mr *lookup_mr(struct rxe_pd *pd, int access, u32 key,return mr; } -int rxe_invalidate_mr (struct rxe_qp *qp,u32 rkey) +static int rxe_invalidate_mr (struct rxe_mr *mr) +{ +if (atomic_read (&mr->num_mw) >0 ) { +pr_warn ("%s: Attempt to invalidate an MR while bound to MWs\n" , + __func__); +return -EINVAL ; + } + +if (unlikely (mr->type !=IB_MR_TYPE_MEM_REG )) { +pr_warn ("%s: mr->type (%d) is wrong type\n" , __func__, mr->type); +return -EINVAL ; + } + + mr->state =RXE_MR_STATE_FREE ; +return 0 ; +} + +int rxe_invalidate_mr_local (struct rxe_qp *qp,u32 lkey) {struct rxe_dev *rxe =to_rdev (qp->ibqp.device);struct rxe_mr *mr;int ret; - mr =rxe_pool_get_index (&rxe->mr_pool, rkey >>8 ); + mr =rxe_pool_get_index (&rxe->mr_pool, lkey >>8 );if (!mr) { -pr_err ("%s: No MR for rkey %#x\n" , __func__, rkey); +pr_err ("%s: No MR for lkey %#x\n" , __func__, lkey); ret = -EINVAL ; goto err; } -if (rkey != mr->rkey) { -pr_err ("%s: rkey (%#x) doesn't match mr->rkey (%#x)\n" , - __func__, rkey, mr->rkey); +if (lkey != mr->lkey) { +pr_err ("%s: lkey (%#x) doesn't match mr->lkey (%#x)\n" , + __func__, lkey, mr->lkey); ret = -EINVAL ; goto err_drop_ref; } -if (atomic_read (&mr->num_mw) >0 ) { -pr_warn ("%s: Attempt to invalidate an MR while bound to MWs\n" , - __func__); + ret =rxe_invalidate_mr (mr); + +err_drop_ref: +rxe_put (mr); +err: +return ret; +} + +int rxe_invalidate_mr_remote (struct rxe_qp *qp,u32 rkey) +{ +struct rxe_dev *rxe =to_rdev (qp->ibqp.device); +struct rxe_mr *mr; +int ret; + + mr =rxe_pool_get_index (&rxe->mr_pool, rkey >>8 ); +if (!mr) { +pr_err ("%s: No MR for rkey %#x\n" , __func__, rkey); ret = -EINVAL ; - goto err_drop_ref; + goto err; } -if (unlikely (mr->type !=IB_MR_TYPE_MEM_REG )) { -pr_warn ("%s: mr->type (%d) is wrong type\n" , __func__, mr->type); +if (rkey != mr->rkey) { +pr_err ("%s: rkey (%#x) doesn't match mr->rkey (%#x)\n" , + __func__, rkey, mr->rkey); ret = -EINVAL ; goto err_drop_ref; } - mr->state =RXE_MR_STATE_FREE ; - ret =0 ; + ret =rxe_invalidate_mr (mr); err_drop_ref:rxe_put (mr);drivers/infiniband/sw/rxe/rxe_req.c +5
-5@@ -541,16 +541,16 @@ static void update_state(struct rxe_qp *qp, struct rxe_pkt_info *pkt)static int rxe_do_local_ops (struct rxe_qp *qp,struct rxe_send_wqe *wqe) {u8 opcode = wqe->wr.opcode; -u32 rkey; +u32 key;int ret;switch (opcode) {case IB_WR_LOCAL_INV : - rkey = wqe->wr.ex.invalidate_rkey; -if (rkey_is_mw (rkey)) - ret =rxe_invalidate_mw (qp, rkey); + key = wqe->wr.ex.invalidate_rkey; +if (rkey_is_mw (key)) + ret =rxe_invalidate_mw (qp, key);else - ret =rxe_invalidate_mr (qp, rkey); + ret =rxe_invalidate_mr_local (qp, key);if (unlikely (ret)) { wqe->status =IB_WC_LOC_QP_OP_ERR ;drivers/infiniband/sw/rxe/rxe_resp.c +1
-2@@ -818,7 +818,7 @@ static int invalidate_rkey(struct rxe_qp *qp, u32 rkey)if (rkey_is_mw (rkey))return rxe_invalidate_mw (qp, rkey);else -return rxe_invalidate_mr (qp, rkey); +return rxe_invalidate_mr_remote (qp, rkey); }/* Executes a new request. A retried request never reach that function (send --
Quoted 2 lines
2.25.1
ping.
Quoted 11 lines · haris iqbal
Quoted 10 lines · Md Haris Iqbal
Currently
rxe_invalidate_mrdoes invalidate for both local ops, and remote ones. This means that MR being invalidated is compared with rkey for both, which is incorrect. For local invalidate, comparison should happen with lkey, and for the remote one, it should happen with rkey.This commit splits the
rxe_invalidate_mrinto local and remote versions. Each of them does comparison the right way as described above (with lkey for local, and rkey for remote).
Quoted 5 lines
Quoted 3 lines
Fixes 3902b429ca14 ("RDMA/rxe: Implement invalidate MW operations")Signed-off-by Md Haris Iqbal <[email protected]>ping.
The commit message makes sense to me - is anyone else working on rxe going to test or review this patch?
There are now many patches for rxe that I would like the rxe community to mutually review/test/ack.
Jason
Bob Pearson
Reviewed-byQuoted 18 lines · Jason Gunthorpe
Quoted 11 lines · haris iqbal
Quoted 10 lines · Md Haris Iqbal
Currently
rxe_invalidate_mrdoes invalidate for both local ops, and remote ones. This means that MR being invalidated is compared with rkey for both, which is incorrect. For local invalidate, comparison should happen with lkey, and for the remote one, it should happen with rkey.This commit splits the
rxe_invalidate_mrinto local and remote versions. Each of them does comparison the right way as described above (with lkey for local, and rkey for remote).Quoted 5 lines
Quoted 3 lines
Fixes 3902b429ca14 ("RDMA/rxe: Implement invalidate MW operations")Signed-off-by Md Haris Iqbal <[email protected]>ping.
Quoted 7 lines
The commit message makes sense to me - is anyone else working on rxe going to test or review this patch?
There are now many patches for rxe that I would like the rxe community to mutually review/test/ack.
Jason
I have reviewed this patch. Haris and I discussed it after v1 and we agreed to this for v2.