#include <linux/sunrpc/svc_xprt.h>
#include <linux/sunrpc/debug.h>
#include <linux/sunrpc/rpc_rdma.h>
+#include <linux/sched.h>
#include <linux/spinlock.h>
#include <rdma/ib_verbs.h>
#include <rdma/rdma_cm.h>
static void rq_cq_reap(struct svcxprt_rdma *xprt);
static void sq_cq_reap(struct svcxprt_rdma *xprt);
-DECLARE_TASKLET(dto_tasklet, dto_tasklet_func, 0UL);
+static DECLARE_TASKLET(dto_tasklet, dto_tasklet_func, 0UL);
static DEFINE_SPINLOCK(dto_lock);
static LIST_HEAD(dto_xprt_q);
.xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
};
-static int rdma_bump_context_cache(struct svcxprt_rdma *xprt)
+/* WR context cache. Created in svc_rdma.c */
+extern struct kmem_cache *svc_rdma_ctxt_cachep;
+
+struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *xprt)
{
- int target;
- int at_least_one = 0;
struct svc_rdma_op_ctxt *ctxt;
- target = min(xprt->sc_ctxt_cnt + xprt->sc_ctxt_bump,
- xprt->sc_ctxt_max);
-
- spin_lock_bh(&xprt->sc_ctxt_lock);
- while (xprt->sc_ctxt_cnt < target) {
- xprt->sc_ctxt_cnt++;
- spin_unlock_bh(&xprt->sc_ctxt_lock);
-
- ctxt = kmalloc(sizeof(*ctxt), GFP_KERNEL);
-
- spin_lock_bh(&xprt->sc_ctxt_lock);
- if (ctxt) {
- at_least_one = 1;
- ctxt->next = xprt->sc_ctxt_head;
- xprt->sc_ctxt_head = ctxt;
- } else {
- /* kmalloc failed...give up for now */
- xprt->sc_ctxt_cnt--;
+ while (1) {
+ ctxt = kmem_cache_alloc(svc_rdma_ctxt_cachep, GFP_KERNEL);
+ if (ctxt)
break;
- }
+ schedule_timeout_uninterruptible(msecs_to_jiffies(500));
}
- spin_unlock_bh(&xprt->sc_ctxt_lock);
- dprintk("svcrdma: sc_ctxt_max=%d, sc_ctxt_cnt=%d\n",
- xprt->sc_ctxt_max, xprt->sc_ctxt_cnt);
- return at_least_one;
+ ctxt->xprt = xprt;
+ INIT_LIST_HEAD(&ctxt->dto_q);
+ ctxt->count = 0;
+ ctxt->frmr = NULL;
+ atomic_inc(&xprt->sc_ctxt_used);
+ return ctxt;
}
-struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *xprt)
+void svc_rdma_unmap_dma(struct svc_rdma_op_ctxt *ctxt)
{
- struct svc_rdma_op_ctxt *ctxt;
-
- while (1) {
- spin_lock_bh(&xprt->sc_ctxt_lock);
- if (unlikely(xprt->sc_ctxt_head == NULL)) {
- /* Try to bump my cache. */
- spin_unlock_bh(&xprt->sc_ctxt_lock);
-
- if (rdma_bump_context_cache(xprt))
- continue;
-
- printk(KERN_INFO "svcrdma: sleeping waiting for "
- "context memory on xprt=%p\n",
- xprt);
- schedule_timeout_uninterruptible(msecs_to_jiffies(500));
- continue;
+ struct svcxprt_rdma *xprt = ctxt->xprt;
+ int i;
+ for (i = 0; i < ctxt->count && ctxt->sge[i].length; i++) {
+ /*
+ * Unmap the DMA addr in the SGE if the lkey matches
+ * the sc_dma_lkey, otherwise, ignore it since it is
+ * an FRMR lkey and will be unmapped later when the
+ * last WR that uses it completes.
+ */
+ if (ctxt->sge[i].lkey == xprt->sc_dma_lkey) {
+ atomic_dec(&xprt->sc_dma_used);
+ ib_dma_unmap_single(xprt->sc_cm_id->device,
+ ctxt->sge[i].addr,
+ ctxt->sge[i].length,
+ ctxt->direction);
}
- ctxt = xprt->sc_ctxt_head;
- xprt->sc_ctxt_head = ctxt->next;
- spin_unlock_bh(&xprt->sc_ctxt_lock);
- ctxt->xprt = xprt;
- INIT_LIST_HEAD(&ctxt->dto_q);
- ctxt->count = 0;
- break;
}
- return ctxt;
}
void svc_rdma_put_context(struct svc_rdma_op_ctxt *ctxt, int free_pages)
for (i = 0; i < ctxt->count; i++)
put_page(ctxt->pages[i]);
- for (i = 0; i < ctxt->count; i++)
- dma_unmap_single(xprt->sc_cm_id->device->dma_device,
- ctxt->sge[i].addr,
- ctxt->sge[i].length,
- ctxt->direction);
- spin_lock_bh(&xprt->sc_ctxt_lock);
- ctxt->next = xprt->sc_ctxt_head;
- xprt->sc_ctxt_head = ctxt;
- spin_unlock_bh(&xprt->sc_ctxt_lock);
+ kmem_cache_free(svc_rdma_ctxt_cachep, ctxt);
+ atomic_dec(&xprt->sc_ctxt_used);
+}
+
+/* Temporary NFS request map cache. Created in svc_rdma.c */
+extern struct kmem_cache *svc_rdma_map_cachep;
+
+/*
+ * Temporary NFS req mappings are shared across all transport
+ * instances. These are short lived and should be bounded by the number
+ * of concurrent server threads * depth of the SQ.
+ */
+struct svc_rdma_req_map *svc_rdma_get_req_map(void)
+{
+ struct svc_rdma_req_map *map;
+ while (1) {
+ map = kmem_cache_alloc(svc_rdma_map_cachep, GFP_KERNEL);
+ if (map)
+ break;
+ schedule_timeout_uninterruptible(msecs_to_jiffies(500));
+ }
+ map->count = 0;
+ map->frmr = NULL;
+ return map;
+}
+
+void svc_rdma_put_req_map(struct svc_rdma_req_map *map)
+{
+ kmem_cache_free(svc_rdma_map_cachep, map);
}
/* ib_cq event handler */
struct svcxprt_rdma *xprt = cq_context;
unsigned long flags;
+ /* Guard against unconditional flush call for destroyed QP */
+ if (atomic_read(&xprt->sc_xprt.xpt_ref.refcount)==0)
+ return;
+
/*
* Set the bit regardless of whether or not it's on the list
* because it may be on the list already due to an SQ
* completion.
- */
+ */
set_bit(RDMAXPRT_RQ_PENDING, &xprt->sc_flags);
/*
*
* Take all completing WC off the CQE and enqueue the associated DTO
* context on the dto_q for the transport.
+ *
+ * Note that caller must hold a transport reference.
*/
static void rq_cq_reap(struct svcxprt_rdma *xprt)
{
ib_req_notify_cq(xprt->sc_rq_cq, IB_CQ_NEXT_COMP);
atomic_inc(&rdma_stat_rq_poll);
- spin_lock_bh(&xprt->sc_rq_dto_lock);
while ((ret = ib_poll_cq(xprt->sc_rq_cq, 1, &wc)) > 0) {
ctxt = (struct svc_rdma_op_ctxt *)(unsigned long)wc.wr_id;
ctxt->wc_status = wc.status;
ctxt->byte_len = wc.byte_len;
+ svc_rdma_unmap_dma(ctxt);
if (wc.status != IB_WC_SUCCESS) {
/* Close the transport */
+ dprintk("svcrdma: transport closing putting ctxt %p\n", ctxt);
set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
svc_rdma_put_context(ctxt, 1);
+ svc_xprt_put(&xprt->sc_xprt);
continue;
}
+ spin_lock_bh(&xprt->sc_rq_dto_lock);
list_add_tail(&ctxt->dto_q, &xprt->sc_rq_dto_q);
+ spin_unlock_bh(&xprt->sc_rq_dto_lock);
+ svc_xprt_put(&xprt->sc_xprt);
}
- spin_unlock_bh(&xprt->sc_rq_dto_lock);
if (ctxt)
atomic_inc(&rdma_stat_rq_prod);
}
/*
+ * Processs a completion context
+ */
+static void process_context(struct svcxprt_rdma *xprt,
+ struct svc_rdma_op_ctxt *ctxt)
+{
+ svc_rdma_unmap_dma(ctxt);
+
+ switch (ctxt->wr_op) {
+ case IB_WR_SEND:
+ if (test_bit(RDMACTXT_F_FAST_UNREG, &ctxt->flags))
+ svc_rdma_put_frmr(xprt, ctxt->frmr);
+ svc_rdma_put_context(ctxt, 1);
+ break;
+
+ case IB_WR_RDMA_WRITE:
+ svc_rdma_put_context(ctxt, 0);
+ break;
+
+ case IB_WR_RDMA_READ:
+ case IB_WR_RDMA_READ_WITH_INV:
+ if (test_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags)) {
+ struct svc_rdma_op_ctxt *read_hdr = ctxt->read_hdr;
+ BUG_ON(!read_hdr);
+ if (test_bit(RDMACTXT_F_FAST_UNREG, &ctxt->flags))
+ svc_rdma_put_frmr(xprt, ctxt->frmr);
+ spin_lock_bh(&xprt->sc_rq_dto_lock);
+ set_bit(XPT_DATA, &xprt->sc_xprt.xpt_flags);
+ list_add_tail(&read_hdr->dto_q,
+ &xprt->sc_read_complete_q);
+ spin_unlock_bh(&xprt->sc_rq_dto_lock);
+ svc_xprt_enqueue(&xprt->sc_xprt);
+ }
+ svc_rdma_put_context(ctxt, 0);
+ break;
+
+ default:
+ printk(KERN_ERR "svcrdma: unexpected completion type, "
+ "opcode=%d\n",
+ ctxt->wr_op);
+ break;
+ }
+}
+
+/*
* Send Queue Completion Handler - potentially called on interrupt context.
+ *
+ * Note that caller must hold a transport reference.
*/
static void sq_cq_reap(struct svcxprt_rdma *xprt)
{
struct ib_cq *cq = xprt->sc_sq_cq;
int ret;
-
if (!test_and_clear_bit(RDMAXPRT_SQ_PENDING, &xprt->sc_flags))
return;
ib_req_notify_cq(xprt->sc_sq_cq, IB_CQ_NEXT_COMP);
atomic_inc(&rdma_stat_sq_poll);
while ((ret = ib_poll_cq(cq, 1, &wc)) > 0) {
- ctxt = (struct svc_rdma_op_ctxt *)(unsigned long)wc.wr_id;
- xprt = ctxt->xprt;
-
if (wc.status != IB_WC_SUCCESS)
/* Close the transport */
set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
atomic_dec(&xprt->sc_sq_count);
wake_up(&xprt->sc_send_wait);
- switch (ctxt->wr_op) {
- case IB_WR_SEND:
- case IB_WR_RDMA_WRITE:
- svc_rdma_put_context(ctxt, 1);
- break;
-
- case IB_WR_RDMA_READ:
- if (test_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags)) {
- set_bit(XPT_DATA, &xprt->sc_xprt.xpt_flags);
- set_bit(RDMACTXT_F_READ_DONE, &ctxt->flags);
- spin_lock_bh(&xprt->sc_read_complete_lock);
- list_add_tail(&ctxt->dto_q,
- &xprt->sc_read_complete_q);
- spin_unlock_bh(&xprt->sc_read_complete_lock);
- svc_xprt_enqueue(&xprt->sc_xprt);
- }
- break;
+ ctxt = (struct svc_rdma_op_ctxt *)(unsigned long)wc.wr_id;
+ if (ctxt)
+ process_context(xprt, ctxt);
- default:
- printk(KERN_ERR "svcrdma: unexpected completion type, "
- "opcode=%d, status=%d\n",
- wc.opcode, wc.status);
- break;
- }
+ svc_xprt_put(&xprt->sc_xprt);
}
if (ctxt)
struct svcxprt_rdma *xprt = cq_context;
unsigned long flags;
+ /* Guard against unconditional flush call for destroyed QP */
+ if (atomic_read(&xprt->sc_xprt.xpt_ref.refcount)==0)
+ return;
+
/*
* Set the bit regardless of whether or not it's on the list
* because it may be on the list already due to an RQ
* completion.
- */
+ */
set_bit(RDMAXPRT_SQ_PENDING, &xprt->sc_flags);
/*
tasklet_schedule(&dto_tasklet);
}
-static void create_context_cache(struct svcxprt_rdma *xprt,
- int ctxt_count, int ctxt_bump, int ctxt_max)
-{
- struct svc_rdma_op_ctxt *ctxt;
- int i;
-
- xprt->sc_ctxt_max = ctxt_max;
- xprt->sc_ctxt_bump = ctxt_bump;
- xprt->sc_ctxt_cnt = 0;
- xprt->sc_ctxt_head = NULL;
- for (i = 0; i < ctxt_count; i++) {
- ctxt = kmalloc(sizeof(*ctxt), GFP_KERNEL);
- if (ctxt) {
- ctxt->next = xprt->sc_ctxt_head;
- xprt->sc_ctxt_head = ctxt;
- xprt->sc_ctxt_cnt++;
- }
- }
-}
-
-static void destroy_context_cache(struct svc_rdma_op_ctxt *ctxt)
-{
- struct svc_rdma_op_ctxt *next;
- if (!ctxt)
- return;
-
- do {
- next = ctxt->next;
- kfree(ctxt);
- ctxt = next;
- } while (next);
-}
-
static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *serv,
int listener)
{
INIT_LIST_HEAD(&cma_xprt->sc_dto_q);
INIT_LIST_HEAD(&cma_xprt->sc_rq_dto_q);
INIT_LIST_HEAD(&cma_xprt->sc_read_complete_q);
+ INIT_LIST_HEAD(&cma_xprt->sc_frmr_q);
init_waitqueue_head(&cma_xprt->sc_send_wait);
spin_lock_init(&cma_xprt->sc_lock);
- spin_lock_init(&cma_xprt->sc_read_complete_lock);
- spin_lock_init(&cma_xprt->sc_ctxt_lock);
spin_lock_init(&cma_xprt->sc_rq_dto_lock);
+ spin_lock_init(&cma_xprt->sc_frmr_q_lock);
cma_xprt->sc_ord = svcrdma_ord;
cma_xprt->sc_max_requests = svcrdma_max_requests;
cma_xprt->sc_sq_depth = svcrdma_max_requests * RPCRDMA_SQ_DEPTH_MULT;
atomic_set(&cma_xprt->sc_sq_count, 0);
+ atomic_set(&cma_xprt->sc_ctxt_used, 0);
- if (!listener) {
- int reqs = cma_xprt->sc_max_requests;
- create_context_cache(cma_xprt,
- reqs << 1, /* starting size */
- reqs, /* bump amount */
- reqs +
- cma_xprt->sc_sq_depth +
- RPCRDMA_MAX_THREADS + 1); /* max */
- if (!cma_xprt->sc_ctxt_head) {
- kfree(cma_xprt);
- return NULL;
- }
- clear_bit(XPT_LISTENER, &cma_xprt->sc_xprt.xpt_flags);
- } else
+ if (listener)
set_bit(XPT_LISTENER, &cma_xprt->sc_xprt.xpt_flags);
return cma_xprt;
struct ib_recv_wr recv_wr, *bad_recv_wr;
struct svc_rdma_op_ctxt *ctxt;
struct page *page;
- unsigned long pa;
+ dma_addr_t pa;
int sge_no;
int buflen;
int ret;
BUG_ON(sge_no >= xprt->sc_max_sge);
page = svc_rdma_get_page();
ctxt->pages[sge_no] = page;
- pa = ib_dma_map_page(xprt->sc_cm_id->device,
- page, 0, PAGE_SIZE,
+ pa = ib_dma_map_single(xprt->sc_cm_id->device,
+ page_address(page), PAGE_SIZE,
DMA_FROM_DEVICE);
+ if (ib_dma_mapping_error(xprt->sc_cm_id->device, pa))
+ goto err_put_ctxt;
+ atomic_inc(&xprt->sc_dma_used);
ctxt->sge[sge_no].addr = pa;
ctxt->sge[sge_no].length = PAGE_SIZE;
- ctxt->sge[sge_no].lkey = xprt->sc_phys_mr->lkey;
+ ctxt->sge[sge_no].lkey = xprt->sc_dma_lkey;
buflen += PAGE_SIZE;
}
ctxt->count = sge_no;
recv_wr.num_sge = ctxt->count;
recv_wr.wr_id = (u64)(unsigned long)ctxt;
+ svc_xprt_get(&xprt->sc_xprt);
ret = ib_post_recv(xprt->sc_qp, &recv_wr, &bad_recv_wr);
- if (ret)
+ if (ret) {
+ svc_rdma_unmap_dma(ctxt);
svc_rdma_put_context(ctxt, 1);
+ svc_xprt_put(&xprt->sc_xprt);
+ }
return ret;
+
+ err_put_ctxt:
+ svc_rdma_put_context(ctxt, 1);
+ return -ENOMEM;
}
/*
* will call the recvfrom method on the listen xprt which will accept the new
* connection.
*/
-static void handle_connect_req(struct rdma_cm_id *new_cma_id)
+static void handle_connect_req(struct rdma_cm_id *new_cma_id, size_t client_ird)
{
struct svcxprt_rdma *listen_xprt = new_cma_id->context;
struct svcxprt_rdma *newxprt;
+ struct sockaddr *sa;
/* Create a new transport */
newxprt = rdma_create_xprt(listen_xprt->sc_xprt.xpt_server, 0);
dprintk("svcrdma: Creating newxprt=%p, cm_id=%p, listenxprt=%p\n",
newxprt, newxprt->sc_cm_id, listen_xprt);
+ /* Save client advertised inbound read limit for use later in accept. */
+ newxprt->sc_ord = client_ird;
+
+ /* Set the local and remote addresses in the transport */
+ sa = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.dst_addr;
+ svc_xprt_set_remote(&newxprt->sc_xprt, sa, svc_addr_len(sa));
+ sa = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.src_addr;
+ svc_xprt_set_local(&newxprt->sc_xprt, sa, svc_addr_len(sa));
+
/*
* Enqueue the new transport on the accept queue of the listening
* transport
case RDMA_CM_EVENT_CONNECT_REQUEST:
dprintk("svcrdma: Connect request on cma_id=%p, xprt = %p, "
"event=%d\n", cma_id, cma_id->context, event->event);
- handle_connect_req(cma_id);
+ handle_connect_req(cma_id,
+ event->param.conn.initiator_depth);
break;
case RDMA_CM_EVENT_ESTABLISHED:
cma_xprt = rdma_create_xprt(serv, 1);
if (!cma_xprt)
- return ERR_PTR(ENOMEM);
+ return ERR_PTR(-ENOMEM);
xprt = &cma_xprt->sc_xprt;
listen_id = rdma_create_id(rdma_listen_handler, cma_xprt, RDMA_PS_TCP);
if (IS_ERR(listen_id)) {
- svc_xprt_put(&cma_xprt->sc_xprt);
- dprintk("svcrdma: rdma_create_id failed = %ld\n",
- PTR_ERR(listen_id));
- return (void *)listen_id;
+ ret = PTR_ERR(listen_id);
+ dprintk("svcrdma: rdma_create_id failed = %d\n", ret);
+ goto err0;
}
+
ret = rdma_bind_addr(listen_id, sa);
if (ret) {
- rdma_destroy_id(listen_id);
- svc_xprt_put(&cma_xprt->sc_xprt);
dprintk("svcrdma: rdma_bind_addr failed = %d\n", ret);
- return ERR_PTR(ret);
+ goto err1;
}
cma_xprt->sc_cm_id = listen_id;
ret = rdma_listen(listen_id, RPCRDMA_LISTEN_BACKLOG);
if (ret) {
- rdma_destroy_id(listen_id);
- svc_xprt_put(&cma_xprt->sc_xprt);
dprintk("svcrdma: rdma_listen failed = %d\n", ret);
- return ERR_PTR(ret);
+ goto err1;
}
/*
svc_xprt_set_local(&cma_xprt->sc_xprt, sa, salen);
return &cma_xprt->sc_xprt;
+
+ err1:
+ rdma_destroy_id(listen_id);
+ err0:
+ kfree(cma_xprt);
+ return ERR_PTR(ret);
+}
+
+static struct svc_rdma_fastreg_mr *rdma_alloc_frmr(struct svcxprt_rdma *xprt)
+{
+ struct ib_mr *mr;
+ struct ib_fast_reg_page_list *pl;
+ struct svc_rdma_fastreg_mr *frmr;
+
+ frmr = kmalloc(sizeof(*frmr), GFP_KERNEL);
+ if (!frmr)
+ goto err;
+
+ mr = ib_alloc_fast_reg_mr(xprt->sc_pd, RPCSVC_MAXPAGES);
+ if (IS_ERR(mr))
+ goto err_free_frmr;
+
+ pl = ib_alloc_fast_reg_page_list(xprt->sc_cm_id->device,
+ RPCSVC_MAXPAGES);
+ if (IS_ERR(pl))
+ goto err_free_mr;
+
+ frmr->mr = mr;
+ frmr->page_list = pl;
+ INIT_LIST_HEAD(&frmr->frmr_list);
+ return frmr;
+
+ err_free_mr:
+ ib_dereg_mr(mr);
+ err_free_frmr:
+ kfree(frmr);
+ err:
+ return ERR_PTR(-ENOMEM);
+}
+
+static void rdma_dealloc_frmr_q(struct svcxprt_rdma *xprt)
+{
+ struct svc_rdma_fastreg_mr *frmr;
+
+ while (!list_empty(&xprt->sc_frmr_q)) {
+ frmr = list_entry(xprt->sc_frmr_q.next,
+ struct svc_rdma_fastreg_mr, frmr_list);
+ list_del_init(&frmr->frmr_list);
+ ib_dereg_mr(frmr->mr);
+ ib_free_fast_reg_page_list(frmr->page_list);
+ kfree(frmr);
+ }
+}
+
+struct svc_rdma_fastreg_mr *svc_rdma_get_frmr(struct svcxprt_rdma *rdma)
+{
+ struct svc_rdma_fastreg_mr *frmr = NULL;
+
+ spin_lock_bh(&rdma->sc_frmr_q_lock);
+ if (!list_empty(&rdma->sc_frmr_q)) {
+ frmr = list_entry(rdma->sc_frmr_q.next,
+ struct svc_rdma_fastreg_mr, frmr_list);
+ list_del_init(&frmr->frmr_list);
+ frmr->map_len = 0;
+ frmr->page_list_len = 0;
+ }
+ spin_unlock_bh(&rdma->sc_frmr_q_lock);
+ if (frmr)
+ return frmr;
+
+ return rdma_alloc_frmr(rdma);
+}
+
+static void frmr_unmap_dma(struct svcxprt_rdma *xprt,
+ struct svc_rdma_fastreg_mr *frmr)
+{
+ int page_no;
+ for (page_no = 0; page_no < frmr->page_list_len; page_no++) {
+ dma_addr_t addr = frmr->page_list->page_list[page_no];
+ if (ib_dma_mapping_error(frmr->mr->device, addr))
+ continue;
+ atomic_dec(&xprt->sc_dma_used);
+ ib_dma_unmap_single(frmr->mr->device, addr, PAGE_SIZE,
+ frmr->direction);
+ }
+}
+
+void svc_rdma_put_frmr(struct svcxprt_rdma *rdma,
+ struct svc_rdma_fastreg_mr *frmr)
+{
+ if (frmr) {
+ frmr_unmap_dma(rdma, frmr);
+ spin_lock_bh(&rdma->sc_frmr_q_lock);
+ BUG_ON(!list_empty(&frmr->frmr_list));
+ list_add(&frmr->frmr_list, &rdma->sc_frmr_q);
+ spin_unlock_bh(&rdma->sc_frmr_q_lock);
+ }
}
/*
struct rdma_conn_param conn_param;
struct ib_qp_init_attr qp_attr;
struct ib_device_attr devattr;
- struct sockaddr *sa;
+ int uninitialized_var(dma_mr_acc);
+ int need_dma_mr;
int ret;
int i;
(size_t)svcrdma_max_requests);
newxprt->sc_sq_depth = RPCRDMA_SQ_DEPTH_MULT * newxprt->sc_max_requests;
- newxprt->sc_ord = min((size_t)devattr.max_qp_rd_atom,
- (size_t)svcrdma_ord);
+ /*
+ * Limit ORD based on client limit, local device limit, and
+ * configured svcrdma limit.
+ */
+ newxprt->sc_ord = min_t(size_t, devattr.max_qp_rd_atom, newxprt->sc_ord);
+ newxprt->sc_ord = min_t(size_t, svcrdma_ord, newxprt->sc_ord);
newxprt->sc_pd = ib_alloc_pd(newxprt->sc_cm_id->device);
if (IS_ERR(newxprt->sc_pd)) {
newxprt->sc_sq_depth = qp_attr.cap.max_send_wr;
newxprt->sc_max_requests = qp_attr.cap.max_recv_wr;
}
- svc_xprt_get(&newxprt->sc_xprt);
newxprt->sc_qp = newxprt->sc_cm_id->qp;
- /* Register all of physical memory */
- newxprt->sc_phys_mr = ib_get_dma_mr(newxprt->sc_pd,
- IB_ACCESS_LOCAL_WRITE |
- IB_ACCESS_REMOTE_WRITE);
- if (IS_ERR(newxprt->sc_phys_mr)) {
- dprintk("svcrdma: Failed to create DMA MR ret=%d\n", ret);
+ /*
+ * Use the most secure set of MR resources based on the
+ * transport type and available memory management features in
+ * the device. Here's the table implemented below:
+ *
+ * Fast Global DMA Remote WR
+ * Reg LKEY MR Access
+ * Sup'd Sup'd Needed Needed
+ *
+ * IWARP N N Y Y
+ * N Y Y Y
+ * Y N Y N
+ * Y Y N -
+ *
+ * IB N N Y N
+ * N Y N -
+ * Y N Y N
+ * Y Y N -
+ *
+ * NB: iWARP requires remote write access for the data sink
+ * of an RDMA_READ. IB does not.
+ */
+ if (devattr.device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS) {
+ newxprt->sc_frmr_pg_list_len =
+ devattr.max_fast_reg_page_list_len;
+ newxprt->sc_dev_caps |= SVCRDMA_DEVCAP_FAST_REG;
+ }
+
+ /*
+ * Determine if a DMA MR is required and if so, what privs are required
+ */
+ switch (rdma_node_get_transport(newxprt->sc_cm_id->device->node_type)) {
+ case RDMA_TRANSPORT_IWARP:
+ newxprt->sc_dev_caps |= SVCRDMA_DEVCAP_READ_W_INV;
+ if (!(newxprt->sc_dev_caps & SVCRDMA_DEVCAP_FAST_REG)) {
+ need_dma_mr = 1;
+ dma_mr_acc =
+ (IB_ACCESS_LOCAL_WRITE |
+ IB_ACCESS_REMOTE_WRITE);
+ } else if (!(devattr.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)) {
+ need_dma_mr = 1;
+ dma_mr_acc = IB_ACCESS_LOCAL_WRITE;
+ } else
+ need_dma_mr = 0;
+ break;
+ case RDMA_TRANSPORT_IB:
+ if (!(devattr.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)) {
+ need_dma_mr = 1;
+ dma_mr_acc = IB_ACCESS_LOCAL_WRITE;
+ } else
+ need_dma_mr = 0;
+ break;
+ default:
goto errout;
}
+ /* Create the DMA MR if needed, otherwise, use the DMA LKEY */
+ if (need_dma_mr) {
+ /* Register all of physical memory */
+ newxprt->sc_phys_mr =
+ ib_get_dma_mr(newxprt->sc_pd, dma_mr_acc);
+ if (IS_ERR(newxprt->sc_phys_mr)) {
+ dprintk("svcrdma: Failed to create DMA MR ret=%d\n",
+ ret);
+ goto errout;
+ }
+ newxprt->sc_dma_lkey = newxprt->sc_phys_mr->lkey;
+ } else
+ newxprt->sc_dma_lkey =
+ newxprt->sc_cm_id->device->local_dma_lkey;
+
/* Post receive buffers */
for (i = 0; i < newxprt->sc_max_requests; i++) {
ret = svc_rdma_post_recv(newxprt);
/* Swap out the handler */
newxprt->sc_cm_id->event_handler = rdma_cma_handler;
+ /*
+ * Arm the CQs for the SQ and RQ before accepting so we can't
+ * miss the first message
+ */
+ ib_req_notify_cq(newxprt->sc_sq_cq, IB_CQ_NEXT_COMP);
+ ib_req_notify_cq(newxprt->sc_rq_cq, IB_CQ_NEXT_COMP);
+
/* Accept Connection */
set_bit(RDMAXPRT_CONN_PENDING, &newxprt->sc_flags);
memset(&conn_param, 0, sizeof conn_param);
dprintk("svcrdma: new connection %p accepted with the following "
"attributes:\n"
- " local_ip : %d.%d.%d.%d\n"
+ " local_ip : %pI4\n"
" local_port : %d\n"
- " remote_ip : %d.%d.%d.%d\n"
+ " remote_ip : %pI4\n"
" remote_port : %d\n"
" max_sge : %d\n"
" sq_depth : %d\n"
" max_requests : %d\n"
" ord : %d\n",
newxprt,
- NIPQUAD(((struct sockaddr_in *)&newxprt->sc_cm_id->
- route.addr.src_addr)->sin_addr.s_addr),
+ &((struct sockaddr_in *)&newxprt->sc_cm_id->
+ route.addr.src_addr)->sin_addr.s_addr,
ntohs(((struct sockaddr_in *)&newxprt->sc_cm_id->
route.addr.src_addr)->sin_port),
- NIPQUAD(((struct sockaddr_in *)&newxprt->sc_cm_id->
- route.addr.dst_addr)->sin_addr.s_addr),
+ &((struct sockaddr_in *)&newxprt->sc_cm_id->
+ route.addr.dst_addr)->sin_addr.s_addr,
ntohs(((struct sockaddr_in *)&newxprt->sc_cm_id->
route.addr.dst_addr)->sin_port),
newxprt->sc_max_sge,
newxprt->sc_max_requests,
newxprt->sc_ord);
- /* Set the local and remote addresses in the transport */
- sa = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.dst_addr;
- svc_xprt_set_remote(&newxprt->sc_xprt, sa, svc_addr_len(sa));
- sa = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.src_addr;
- svc_xprt_set_local(&newxprt->sc_xprt, sa, svc_addr_len(sa));
-
- ib_req_notify_cq(newxprt->sc_sq_cq, IB_CQ_NEXT_COMP);
- ib_req_notify_cq(newxprt->sc_rq_cq, IB_CQ_NEXT_COMP);
return &newxprt->sc_xprt;
errout:
dprintk("svcrdma: failure accepting new connection rc=%d.\n", ret);
/* Take a reference in case the DTO handler runs */
svc_xprt_get(&newxprt->sc_xprt);
- if (newxprt->sc_qp && !IS_ERR(newxprt->sc_qp)) {
+ if (newxprt->sc_qp && !IS_ERR(newxprt->sc_qp))
ib_destroy_qp(newxprt->sc_qp);
- svc_xprt_put(&newxprt->sc_xprt);
- }
rdma_destroy_id(newxprt->sc_cm_id);
/* This call to put will destroy the transport */
svc_xprt_put(&newxprt->sc_xprt);
}
/*
- * When connected, an svc_xprt has at least three references:
- *
- * - A reference held by the QP. We still hold that here because this
- * code deletes the QP and puts the reference.
+ * When connected, an svc_xprt has at least two references:
*
* - A reference held by the cm_id between the ESTABLISHED and
* DISCONNECTED events. If the remote peer disconnected first, this
* - A reference held by the svc_recv code that called this function
* as part of close processing.
*
- * At a minimum two references should still be held.
+ * At a minimum one references should still be held.
*/
static void svc_rdma_detach(struct svc_xprt *xprt)
{
/* Disconnect and flush posted WQE */
rdma_disconnect(rdma->sc_cm_id);
-
- /* Destroy the QP if present (not a listener) */
- if (rdma->sc_qp && !IS_ERR(rdma->sc_qp)) {
- ib_destroy_qp(rdma->sc_qp);
- svc_xprt_put(xprt);
- }
-
- /* Destroy the CM ID */
- rdma_destroy_id(rdma->sc_cm_id);
}
-static void svc_rdma_free(struct svc_xprt *xprt)
+static void __svc_rdma_free(struct work_struct *work)
{
- struct svcxprt_rdma *rdma = (struct svcxprt_rdma *)xprt;
+ struct svcxprt_rdma *rdma =
+ container_of(work, struct svcxprt_rdma, sc_work);
dprintk("svcrdma: svc_rdma_free(%p)\n", rdma);
+
/* We should only be called from kref_put */
- BUG_ON(atomic_read(&xprt->xpt_ref.refcount) != 0);
+ BUG_ON(atomic_read(&rdma->sc_xprt.xpt_ref.refcount) != 0);
+
+ /*
+ * Destroy queued, but not processed read completions. Note
+ * that this cleanup has to be done before destroying the
+ * cm_id because the device ptr is needed to unmap the dma in
+ * svc_rdma_put_context.
+ */
+ while (!list_empty(&rdma->sc_read_complete_q)) {
+ struct svc_rdma_op_ctxt *ctxt;
+ ctxt = list_entry(rdma->sc_read_complete_q.next,
+ struct svc_rdma_op_ctxt,
+ dto_q);
+ list_del_init(&ctxt->dto_q);
+ svc_rdma_put_context(ctxt, 1);
+ }
+
+ /* Destroy queued, but not processed recv completions */
+ while (!list_empty(&rdma->sc_rq_dto_q)) {
+ struct svc_rdma_op_ctxt *ctxt;
+ ctxt = list_entry(rdma->sc_rq_dto_q.next,
+ struct svc_rdma_op_ctxt,
+ dto_q);
+ list_del_init(&ctxt->dto_q);
+ svc_rdma_put_context(ctxt, 1);
+ }
+
+ /* Warn if we leaked a resource or under-referenced */
+ WARN_ON(atomic_read(&rdma->sc_ctxt_used) != 0);
+ WARN_ON(atomic_read(&rdma->sc_dma_used) != 0);
+
+ /* De-allocate fastreg mr */
+ rdma_dealloc_frmr_q(rdma);
+
+ /* Destroy the QP if present (not a listener) */
+ if (rdma->sc_qp && !IS_ERR(rdma->sc_qp))
+ ib_destroy_qp(rdma->sc_qp);
+
if (rdma->sc_sq_cq && !IS_ERR(rdma->sc_sq_cq))
ib_destroy_cq(rdma->sc_sq_cq);
if (rdma->sc_pd && !IS_ERR(rdma->sc_pd))
ib_dealloc_pd(rdma->sc_pd);
- destroy_context_cache(rdma->sc_ctxt_head);
+ /* Destroy the CM ID */
+ rdma_destroy_id(rdma->sc_cm_id);
+
kfree(rdma);
}
+static void svc_rdma_free(struct svc_xprt *xprt)
+{
+ struct svcxprt_rdma *rdma =
+ container_of(xprt, struct svcxprt_rdma, sc_xprt);
+ INIT_WORK(&rdma->sc_work, __svc_rdma_free);
+ schedule_work(&rdma->sc_work);
+}
+
static int svc_rdma_has_wspace(struct svc_xprt *xprt)
{
struct svcxprt_rdma *rdma =
return 1;
}
+/*
+ * Attempt to register the kvec representing the RPC memory with the
+ * device.
+ *
+ * Returns:
+ * NULL : The device does not support fastreg or there were no more
+ * fastreg mr.
+ * frmr : The kvec register request was successfully posted.
+ * <0 : An error was encountered attempting to register the kvec.
+ */
+int svc_rdma_fastreg(struct svcxprt_rdma *xprt,
+ struct svc_rdma_fastreg_mr *frmr)
+{
+ struct ib_send_wr fastreg_wr;
+ u8 key;
+
+ /* Bump the key */
+ key = (u8)(frmr->mr->lkey & 0x000000FF);
+ ib_update_fast_reg_key(frmr->mr, ++key);
+
+ /* Prepare FASTREG WR */
+ memset(&fastreg_wr, 0, sizeof fastreg_wr);
+ fastreg_wr.opcode = IB_WR_FAST_REG_MR;
+ fastreg_wr.send_flags = IB_SEND_SIGNALED;
+ fastreg_wr.wr.fast_reg.iova_start = (unsigned long)frmr->kva;
+ fastreg_wr.wr.fast_reg.page_list = frmr->page_list;
+ fastreg_wr.wr.fast_reg.page_list_len = frmr->page_list_len;
+ fastreg_wr.wr.fast_reg.page_shift = PAGE_SHIFT;
+ fastreg_wr.wr.fast_reg.length = frmr->map_len;
+ fastreg_wr.wr.fast_reg.access_flags = frmr->access_flags;
+ fastreg_wr.wr.fast_reg.rkey = frmr->mr->lkey;
+ return svc_rdma_send(xprt, &fastreg_wr);
+}
+
int svc_rdma_send(struct svcxprt_rdma *xprt, struct ib_send_wr *wr)
{
- struct ib_send_wr *bad_wr;
+ struct ib_send_wr *bad_wr, *n_wr;
+ int wr_count;
+ int i;
int ret;
if (test_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags))
return -ENOTCONN;
BUG_ON(wr->send_flags != IB_SEND_SIGNALED);
- BUG_ON(((struct svc_rdma_op_ctxt *)(unsigned long)wr->wr_id)->wr_op !=
- wr->opcode);
+ wr_count = 1;
+ for (n_wr = wr->next; n_wr; n_wr = n_wr->next)
+ wr_count++;
+
/* If the SQ is full, wait until an SQ entry is available */
while (1) {
spin_lock_bh(&xprt->sc_lock);
- if (xprt->sc_sq_depth == atomic_read(&xprt->sc_sq_count)) {
+ if (xprt->sc_sq_depth < atomic_read(&xprt->sc_sq_count) + wr_count) {
spin_unlock_bh(&xprt->sc_lock);
atomic_inc(&rdma_stat_sq_starve);
return 0;
continue;
}
- /* Bumped used SQ WR count and post */
+ /* Take a transport ref for each WR posted */
+ for (i = 0; i < wr_count; i++)
+ svc_xprt_get(&xprt->sc_xprt);
+
+ /* Bump used SQ WR count and post */
+ atomic_add(wr_count, &xprt->sc_sq_count);
ret = ib_post_send(xprt->sc_qp, wr, &bad_wr);
- if (!ret)
- atomic_inc(&xprt->sc_sq_count);
- else
+ if (ret) {
+ set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
+ atomic_sub(wr_count, &xprt->sc_sq_count);
+ for (i = 0; i < wr_count; i ++)
+ svc_xprt_put(&xprt->sc_xprt);
dprintk("svcrdma: failed to post SQ WR rc=%d, "
"sc_sq_count=%d, sc_sq_depth=%d\n",
ret, atomic_read(&xprt->sc_sq_count),
xprt->sc_sq_depth);
+ }
spin_unlock_bh(&xprt->sc_lock);
+ if (ret)
+ wake_up(&xprt->sc_send_wait);
break;
}
return ret;
}
-int svc_rdma_send_error(struct svcxprt_rdma *xprt, struct rpcrdma_msg *rmsgp,
- enum rpcrdma_errcode err)
+void svc_rdma_send_error(struct svcxprt_rdma *xprt, struct rpcrdma_msg *rmsgp,
+ enum rpcrdma_errcode err)
{
struct ib_send_wr err_wr;
struct ib_sge sge;
length = svc_rdma_xdr_encode_error(xprt, rmsgp, err, va);
/* Prepare SGE for local address */
- sge.addr = ib_dma_map_page(xprt->sc_cm_id->device,
- p, 0, PAGE_SIZE, DMA_FROM_DEVICE);
- sge.lkey = xprt->sc_phys_mr->lkey;
+ sge.addr = ib_dma_map_single(xprt->sc_cm_id->device,
+ page_address(p), PAGE_SIZE, DMA_FROM_DEVICE);
+ if (ib_dma_mapping_error(xprt->sc_cm_id->device, sge.addr)) {
+ put_page(p);
+ return;
+ }
+ atomic_inc(&xprt->sc_dma_used);
+ sge.lkey = xprt->sc_dma_lkey;
sge.length = length;
ctxt = svc_rdma_get_context(xprt);
/* Post It */
ret = svc_rdma_send(xprt, &err_wr);
if (ret) {
- dprintk("svcrdma: Error posting send = %d\n", ret);
+ dprintk("svcrdma: Error %d posting send for protocol error\n",
+ ret);
+ ib_dma_unmap_single(xprt->sc_cm_id->device,
+ sge.addr, PAGE_SIZE,
+ DMA_FROM_DEVICE);
svc_rdma_put_context(ctxt, 1);
}
-
- return ret;
}
git://ftp.safe.ca / safe / jmp / linux-2.6 / blobdiff