RDMA/iser: Remove support for FMR memory registration
FMR is not supported on most recent RDMA devices (that use fast memory
registration mechanism). Also, FMR was recently removed from NFS/RDMA
ULP.
Link: https://lore.kernel.org/r/1-v3-f58e6669d5d3+2cf-fmr_removal_jgg@mellanox.com
Signed-off-by: Israel Rukshin <israelr@mellanox.com>
Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
diff --git a/drivers/infiniband/ulp/iser/iser_verbs.c b/drivers/infiniband/ulp/iser/iser_verbs.c
index 127887c..c1f44c4 100644
--- a/drivers/infiniband/ulp/iser/iser_verbs.c
+++ b/drivers/infiniband/ulp/iser/iser_verbs.c
@@ -68,11 +68,12 @@ static void iser_event_handler(struct ib_event_handler *handler,
static int iser_create_device_ib_res(struct iser_device *device)
{
struct ib_device *ib_dev = device->ib_device;
- int ret, i, max_cqe;
+ int i, max_cqe;
- ret = iser_assign_reg_ops(device);
- if (ret)
- return ret;
+ if (!(ib_dev->attrs.device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS)) {
+ iser_err("IB device does not support memory registrations\n");
+ return -1;
+ }
device->comps_used = min_t(int, num_online_cpus(),
ib_dev->num_comp_vectors);
@@ -147,96 +148,6 @@ static void iser_free_device_ib_res(struct iser_device *device)
device->pd = NULL;
}
-/**
- * iser_alloc_fmr_pool - Creates FMR pool and page_vector
- * @ib_conn: connection RDMA resources
- * @cmds_max: max number of SCSI commands for this connection
- * @size: max number of pages per map request
- *
- * Return: 0 on success, or errno code on failure
- */
-int iser_alloc_fmr_pool(struct ib_conn *ib_conn,
- unsigned cmds_max,
- unsigned int size)
-{
- struct iser_device *device = ib_conn->device;
- struct iser_fr_pool *fr_pool = &ib_conn->fr_pool;
- struct iser_page_vec *page_vec;
- struct iser_fr_desc *desc;
- struct ib_fmr_pool *fmr_pool;
- struct ib_fmr_pool_param params;
- int ret;
-
- INIT_LIST_HEAD(&fr_pool->list);
- spin_lock_init(&fr_pool->lock);
-
- desc = kzalloc(sizeof(*desc), GFP_KERNEL);
- if (!desc)
- return -ENOMEM;
-
- page_vec = kmalloc(sizeof(*page_vec) + (sizeof(u64) * size),
- GFP_KERNEL);
- if (!page_vec) {
- ret = -ENOMEM;
- goto err_frpl;
- }
-
- page_vec->pages = (u64 *)(page_vec + 1);
-
- params.page_shift = ilog2(SZ_4K);
- params.max_pages_per_fmr = size;
- /* make the pool size twice the max number of SCSI commands *
- * the ML is expected to queue, watermark for unmap at 50% */
- params.pool_size = cmds_max * 2;
- params.dirty_watermark = cmds_max;
- params.cache = 0;
- params.flush_function = NULL;
- params.access = (IB_ACCESS_LOCAL_WRITE |
- IB_ACCESS_REMOTE_WRITE |
- IB_ACCESS_REMOTE_READ);
-
- fmr_pool = ib_create_fmr_pool(device->pd, ¶ms);
- if (IS_ERR(fmr_pool)) {
- ret = PTR_ERR(fmr_pool);
- iser_err("FMR allocation failed, err %d\n", ret);
- goto err_fmr;
- }
-
- desc->rsc.page_vec = page_vec;
- desc->rsc.fmr_pool = fmr_pool;
- list_add(&desc->list, &fr_pool->list);
-
- return 0;
-
-err_fmr:
- kfree(page_vec);
-err_frpl:
- kfree(desc);
-
- return ret;
-}
-
-/**
- * iser_free_fmr_pool - releases the FMR pool and page vec
- * @ib_conn: connection RDMA resources
- */
-void iser_free_fmr_pool(struct ib_conn *ib_conn)
-{
- struct iser_fr_pool *fr_pool = &ib_conn->fr_pool;
- struct iser_fr_desc *desc;
-
- desc = list_first_entry(&fr_pool->list,
- struct iser_fr_desc, list);
- list_del(&desc->list);
-
- iser_info("freeing conn %p fmr pool %p\n",
- ib_conn, desc->rsc.fmr_pool);
-
- ib_destroy_fmr_pool(desc->rsc.fmr_pool);
- kfree(desc->rsc.page_vec);
- kfree(desc);
-}
-
static struct iser_fr_desc *
iser_create_fastreg_desc(struct iser_device *device,
struct ib_pd *pd,
@@ -667,13 +578,12 @@ iser_calc_scsi_params(struct iser_conn *iser_conn,
u32 max_num_sg;
/*
- * FRs without SG_GAPS or FMRs can only map up to a (device) page per
- * entry, but if the first entry is misaligned we'll end up using two
- * entries (head and tail) for a single page worth data, so one
- * additional entry is required.
+ * FRs without SG_GAPS can only map up to a (device) page per entry,
+ * but if the first entry is misaligned we'll end up using two entries
+ * (head and tail) for a single page worth data, so one additional
+ * entry is required.
*/
- if ((attr->device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS) &&
- (attr->device_cap_flags & IB_DEVICE_SG_GAPS_REG))
+ if (attr->device_cap_flags & IB_DEVICE_SG_GAPS_REG)
reserved_mr_pages = 0;
else
reserved_mr_pages = 1;
@@ -684,14 +594,8 @@ iser_calc_scsi_params(struct iser_conn *iser_conn,
max_num_sg = attr->max_fast_reg_page_list_len;
sg_tablesize = DIV_ROUND_UP(max_sectors * SECTOR_SIZE, SZ_4K);
- if (attr->device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS)
- sup_sg_tablesize =
- min_t(
- uint, ISCSI_ISER_MAX_SG_TABLESIZE,
- max_num_sg - reserved_mr_pages);
- else
- sup_sg_tablesize = ISCSI_ISER_MAX_SG_TABLESIZE;
-
+ sup_sg_tablesize = min_t(uint, ISCSI_ISER_MAX_SG_TABLESIZE,
+ max_num_sg - reserved_mr_pages);
iser_conn->scsi_sg_tablesize = min(sg_tablesize, sup_sg_tablesize);
iser_conn->pages_per_mr =
iser_conn->scsi_sg_tablesize + reserved_mr_pages;
@@ -755,7 +659,7 @@ static void iser_route_handler(struct rdma_cm_id *cma_id)
struct iser_cm_hdr req_hdr;
struct iser_conn *iser_conn = (struct iser_conn *)cma_id->context;
struct ib_conn *ib_conn = &iser_conn->ib_conn;
- struct iser_device *device = ib_conn->device;
+ struct ib_device *ib_dev = ib_conn->device->ib_device;
if (iser_conn->state != ISER_CONN_PENDING)
/* bailout */
@@ -766,14 +670,14 @@ static void iser_route_handler(struct rdma_cm_id *cma_id)
goto failure;
memset(&conn_param, 0, sizeof conn_param);
- conn_param.responder_resources = device->ib_device->attrs.max_qp_rd_atom;
+ conn_param.responder_resources = ib_dev->attrs.max_qp_rd_atom;
conn_param.initiator_depth = 1;
conn_param.retry_count = 7;
conn_param.rnr_retry_count = 6;
memset(&req_hdr, 0, sizeof(req_hdr));
req_hdr.flags = ISER_ZBVA_NOT_SUP;
- if (!device->remote_inv_sup)
+ if (!iser_always_reg)
req_hdr.flags |= ISER_SEND_W_INV_NOT_SUP;
conn_param.private_data = (void *)&req_hdr;
conn_param.private_data_len = sizeof(struct iser_cm_hdr);