/**
* ib_alloc_pd - Allocates an unused protection domain.
* @device: The device on which to allocate the protection domain.
*
* A protection domain object provides an association between QPs, shared
* receive queues, address handles, memory regions, and memory windows.
*
* Every PD has a local_dma_lkey which can be used as the lkey value for local
* memory operations.
*/
struct ib_pd *ib_alloc_pd(struct ib_device *device)
{
struct ib_pd *pd;
pd = device->alloc_pd(device, NULL, NULL);
if (IS_ERR(pd))
return pd;
pd->device = device;
pd->uobject = NULL;
pd->local_mr = NULL;
atomic_set(&pd->usecnt, 0);
if (device->attrs.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)
pd->local_dma_lkey = device->local_dma_lkey;
else {
struct ib_mr *mr;
mr = ib_get_dma_mr(pd, IB_ACCESS_LOCAL_WRITE);
if (IS_ERR(mr)) {
ib_dealloc_pd(pd);
return (struct ib_pd *)mr;
}
pd->local_mr = mr;
pd->local_dma_lkey = pd->local_mr->lkey;
}
return pd;
}
static void
roq_eth_cleanup_ofa(struct roq_eth_priv *vdev)
{
int i;
if (vdev->send_cq)
ib_destroy_cq(vdev->send_cq);
if (vdev->recv_cq && vdev->recv_cq != vdev->send_cq)
ib_destroy_cq(vdev->recv_cq);
if (vdev->qps) {
for (i = 0; i < vdev->part_size; i++)
if (vdev->qps[i])
ib_destroy_qp(vdev->qps[i]);
kfree(vdev->qps);
}
if (vdev->qps_rem) {
for (i = 0; i < vdev->rem_part_size; i++)
if (vdev->qps_rem[i])
ib_destroy_qp(vdev->qps_rem[i]);
kfree(vdev->qps_rem);
}
if (vdev->kpd)
ib_dealloc_pd(vdev->kpd);
vdev->qps = vdev->qps_rem = NULL;
vdev->recv_cq = vdev->send_cq = NULL;
vdev->kpd = NULL;
return;
}
/* ib_create_srq ib_destroy_srq */
int srq_1(
IN struct config_t *config,
IN struct resources *res,
IN struct ib_client *test_client)
{
struct ib_pd *pd = ERR_PTR(-EINVAL);
struct ib_srq *srq = ERR_PTR(-EINVAL);
int result = -1;
int rc;
TEST_CASE("ib_create_srq ib_destroy_srq");
pd = ib_alloc_pd(res->device);
CHECK_PTR("ib_alloc_pd", !IS_ERR(pd), goto cleanup);
res->attributes.srq_init_attr.event_handler = srq_event_handler;
init_srq_cap(config, res, &res->attributes.srq_init_attr);
srq = ib_create_srq(pd, &res->attributes.srq_init_attr);
CHECK_PTR("ib_create_srq", !IS_ERR(srq), goto cleanup);
rc = ib_destroy_srq(srq);
CHECK_VALUE("ib_destroy_srq", rc, 0, goto cleanup);
PASSED;
result = 0;
cleanup:
if (!IS_ERR(pd)) {
rc = ib_dealloc_pd(pd);
CHECK_VALUE("ib_dealloc_pd", rc, 0, return -1);
}
static void ib_sock_mem_fini_common(struct IB_SOCK *sock)
{
if (sock->is_mem.ism_mr)
ib_dereg_mr(sock->is_mem.ism_mr);
if (sock->is_mem.ism_pd)
ib_dealloc_pd(sock->is_mem.ism_pd);
}
static int get_port_caps(struct mlx5_ib_dev *dev)
{
struct ib_device_attr *dprops = NULL;
struct ib_port_attr *pprops = NULL;
int err = -ENOMEM;
int port;
struct ib_udata uhw = {.inlen = 0, .outlen = 0};
pprops = kmalloc(sizeof(*pprops), GFP_KERNEL);
if (!pprops)
goto out;
dprops = kmalloc(sizeof(*dprops), GFP_KERNEL);
if (!dprops)
goto out;
err = mlx5_ib_query_device(&dev->ib_dev, dprops, &uhw);
if (err) {
mlx5_ib_warn(dev, "query_device failed %d\n", err);
goto out;
}
for (port = 1; port <= MLX5_CAP_GEN(dev->mdev, num_ports); port++) {
err = mlx5_ib_query_port(&dev->ib_dev, port, pprops);
if (err) {
mlx5_ib_warn(dev, "query_port %d failed %d\n",
port, err);
break;
}
dev->mdev->port_caps[port - 1].pkey_table_len =
dprops->max_pkeys;
dev->mdev->port_caps[port - 1].gid_table_len =
pprops->gid_tbl_len;
mlx5_ib_dbg(dev, "pkey_table_len %d, gid_table_len %d\n",
dprops->max_pkeys, pprops->gid_tbl_len);
}
out:
kfree(pprops);
kfree(dprops);
return err;
}
static void destroy_umrc_res(struct mlx5_ib_dev *dev)
{
int err;
err = mlx5_mr_cache_cleanup(dev);
if (err)
mlx5_ib_warn(dev, "mr cache cleanup failed\n");
mlx5_ib_destroy_qp(dev->umrc.qp);
ib_destroy_cq(dev->umrc.cq);
ib_dealloc_pd(dev->umrc.pd);
}
/**
* iser_create_device_ib_res - creates Protection Domain (PD), Completion
* Queue (CQ), DMA Memory Region (DMA MR) with the device associated with
* the adapator.
*
* returns 0 on success, -1 on failure
*/
static int iser_create_device_ib_res(struct iser_device *device)
{
device->pd = ib_alloc_pd(device->ib_device);
if (IS_ERR(device->pd))
goto pd_err;
device->rx_cq = ib_create_cq(device->ib_device,
iser_cq_callback,
iser_cq_event_callback,
(void *)device,
ISER_MAX_RX_CQ_LEN, 0);
if (IS_ERR(device->rx_cq))
goto rx_cq_err;
device->tx_cq = ib_create_cq(device->ib_device,
NULL, iser_cq_event_callback,
(void *)device,
ISER_MAX_TX_CQ_LEN, 0);
if (IS_ERR(device->tx_cq))
goto tx_cq_err;
if (ib_req_notify_cq(device->rx_cq, IB_CQ_NEXT_COMP))
goto cq_arm_err;
tasklet_init(&device->cq_tasklet,
iser_cq_tasklet_fn,
(unsigned long)device);
device->mr = ib_get_dma_mr(device->pd, IB_ACCESS_LOCAL_WRITE |
IB_ACCESS_REMOTE_WRITE |
IB_ACCESS_REMOTE_READ);
if (IS_ERR(device->mr))
goto dma_mr_err;
INIT_IB_EVENT_HANDLER(&device->event_handler, device->ib_device,
iser_event_handler);
if (ib_register_event_handler(&device->event_handler))
goto handler_err;
return 0;
handler_err:
ib_dereg_mr(device->mr);
dma_mr_err:
tasklet_kill(&device->cq_tasklet);
cq_arm_err:
ib_destroy_cq(device->tx_cq);
tx_cq_err:
ib_destroy_cq(device->rx_cq);
rx_cq_err:
ib_dealloc_pd(device->pd);
pd_err:
iser_err("failed to allocate an IB resource\n");
return -1;
}
/*
* Open and initialize an Interface Adapter.
* o initializes fields of struct rpcrdma_ia, including
* interface and provider attributes and protection zone.
*/
int
rpcrdma_ia_open(struct rpcrdma_xprt *xprt, struct sockaddr *addr, int memreg)
{
struct rpcrdma_ia *ia = &xprt->rx_ia;
int rc;
ia->ri_id = rpcrdma_create_id(xprt, ia, addr);
if (IS_ERR(ia->ri_id)) {
rc = PTR_ERR(ia->ri_id);
goto out1;
}
ia->ri_device = ia->ri_id->device;
ia->ri_pd = ib_alloc_pd(ia->ri_device, 0);
if (IS_ERR(ia->ri_pd)) {
rc = PTR_ERR(ia->ri_pd);
pr_err("rpcrdma: ib_alloc_pd() returned %d\n", rc);
goto out2;
}
switch (memreg) {
case RPCRDMA_FRMR:
if (frwr_is_supported(ia)) {
ia->ri_ops = &rpcrdma_frwr_memreg_ops;
break;
}
/*FALLTHROUGH*/
case RPCRDMA_MTHCAFMR:
if (fmr_is_supported(ia)) {
ia->ri_ops = &rpcrdma_fmr_memreg_ops;
break;
}
/*FALLTHROUGH*/
default:
pr_err("rpcrdma: Unsupported memory registration mode: %d\n",
memreg);
rc = -EINVAL;
goto out3;
}
return 0;
out3:
ib_dealloc_pd(ia->ri_pd);
ia->ri_pd = NULL;
out2:
rpcrdma_destroy_id(ia->ri_id);
ia->ri_id = NULL;
out1:
return rc;
}
static void iser_free_device_ib_res(struct iser_device *device)
{
BUG_ON(device->mr == NULL);
tasklet_kill(&device->cq_tasklet);
(void)ib_dereg_mr(device->mr);
(void)ib_destroy_cq(device->cq);
(void)ib_dealloc_pd(device->pd);
device->mr = NULL;
device->cq = NULL;
device->pd = NULL;
}
/*
* Clean up/close an IA.
* o if event handles and PD have been initialized, free them.
* o close the IA
*/
void
rpcrdma_ia_close(struct rpcrdma_ia *ia)
{
dprintk("RPC: %s: entering\n", __func__);
if (ia->ri_id != NULL && !IS_ERR(ia->ri_id)) {
if (ia->ri_id->qp)
rdma_destroy_qp(ia->ri_id);
rpcrdma_destroy_id(ia->ri_id);
ia->ri_id = NULL;
}
/* If the pd is still busy, xprtrdma missed freeing a resource */
if (ia->ri_pd && !IS_ERR(ia->ri_pd))
ib_dealloc_pd(ia->ri_pd);
}
static void isert_device_release(struct isert_device *isert_dev)
{
int err, i;
TRACE_ENTRY();
lockdep_assert_held(&dev_list_mutex);
isert_dev_list_remove(isert_dev); /* remove from global list */
for (i = 0; i < isert_dev->num_cqs; ++i) {
struct isert_cq *cq_desc = &isert_dev->cq_desc[i];
#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 22)
/*
* cancel_work_sync() was introduced in 2.6.22. We can
* only wait until all scheduled work is done.
*/
flush_workqueue(cq_desc->cq_workqueue);
#else
cancel_work_sync(&cq_desc->cq_comp_work);
#endif
err = ib_destroy_cq(cq_desc->cq);
if (unlikely(err))
pr_err("Failed to destroy cq, err:%d\n", err);
destroy_workqueue(cq_desc->cq_workqueue);
}
err = ib_dereg_mr(isert_dev->mr);
if (unlikely(err))
pr_err("Failed to destroy mr, err:%d\n", err);
err = ib_dealloc_pd(isert_dev->pd);
if (unlikely(err))
pr_err("Failed to destroy pd, err:%d\n", err);
vfree(isert_dev->cq_desc);
isert_dev->cq_desc = NULL;
kfree(isert_dev->cq_qps);
isert_dev->cq_qps = NULL;
kfree(isert_dev);
TRACE_EXIT();
}
static void verbs_remove_device (struct ib_device *dev)
{
printk (KERN_INFO "IB remove device called. Name = %s\n", dev->name);
if (ah)
ib_destroy_ah (ah);
if (qp)
ib_destroy_qp (qp);
if (send_cq)
ib_destroy_cq (send_cq);
if (recv_cq)
ib_destroy_cq (recv_cq);
if (mr)
ib_dereg_mr (mr);
if (pd)
ib_dealloc_pd (pd);
}
static void rdma_destroy_trans(struct p9_trans_rdma *rdma)
{
if (!rdma)
return;
if (rdma->qp && !IS_ERR(rdma->qp))
ib_destroy_qp(rdma->qp);
if (rdma->pd && !IS_ERR(rdma->pd))
ib_dealloc_pd(rdma->pd);
if (rdma->cq && !IS_ERR(rdma->cq))
ib_free_cq(rdma->cq);
if (rdma->cm_id && !IS_ERR(rdma->cm_id))
rdma_destroy_id(rdma->cm_id);
kfree(rdma);
}
/*
* Clean up/close an IA.
* o if event handles and PD have been initialized, free them.
* o close the IA
*/
void
rpcrdma_ia_close(struct rpcrdma_ia *ia)
{
int rc;
dprintk("RPC: %s: entering\n", __func__);
if (ia->ri_bind_mem != NULL) {
rc = ib_dereg_mr(ia->ri_bind_mem);
dprintk("RPC: %s: ib_dereg_mr returned %i\n",
__func__, rc);
}
if (ia->ri_id != NULL && !IS_ERR(ia->ri_id) && ia->ri_id->qp)
rdma_destroy_qp(ia->ri_id);
if (ia->ri_pd != NULL && !IS_ERR(ia->ri_pd)) {
rc = ib_dealloc_pd(ia->ri_pd);
dprintk("RPC: %s: ib_dealloc_pd returned %i\n",
__func__, rc);
}
if (ia->ri_id != NULL && !IS_ERR(ia->ri_id))
rdma_destroy_id(ia->ri_id);
}
static int krping_setup_qp(struct krping_cb *cb, struct rdma_cm_id *cm_id)
{
int ret;
cb->pd = ib_alloc_pd(cm_id->device);
if (IS_ERR(cb->pd)) {
log(LOG_ERR, "ib_alloc_pd failed\n");
return PTR_ERR(cb->pd);
}
DEBUG_LOG(PFX "created pd %p\n", cb->pd);
cb->cq = ib_create_cq(cm_id->device, krping_cq_event_handler, NULL,
cb, cb->txdepth * 2, 0);
if (IS_ERR(cb->cq)) {
log(LOG_ERR, "ib_create_cq failed\n");
ret = PTR_ERR(cb->cq);
goto err1;
}
DEBUG_LOG(PFX "created cq %p\n", cb->cq);
if (!cb->wlat && !cb->rlat && !cb->bw) {
ret = ib_req_notify_cq(cb->cq, IB_CQ_NEXT_COMP);
if (ret) {
log(LOG_ERR, "ib_create_cq failed\n");
goto err2;
}
}
ret = krping_create_qp(cb);
if (ret) {
log(LOG_ERR, "krping_create_qp failed: %d\n", ret);
goto err2;
}
DEBUG_LOG(PFX "created qp %p\n", cb->qp);
return 0;
err2:
ib_destroy_cq(cb->cq);
err1:
ib_dealloc_pd(cb->pd);
return ret;
}
ssize_t ib_uverbs_dealloc_pd(struct ib_uverbs_file *file,
const char __user *buf,
int in_len, int out_len)
{
struct ib_uverbs_dealloc_pd cmd;
struct ib_pd *pd;
struct ib_uobject *uobj;
int ret = -EINVAL;
if (copy_from_user(&cmd, buf, sizeof cmd))
return -EFAULT;
mutex_lock(&ib_uverbs_idr_mutex);
pd = idr_find(&ib_uverbs_pd_idr, cmd.pd_handle);
if (!pd || pd->uobject->context != file->ucontext)
goto out;
uobj = pd->uobject;
ret = ib_dealloc_pd(pd);
if (ret)
goto out;
idr_remove(&ib_uverbs_pd_idr, cmd.pd_handle);
mutex_lock(&file->mutex);
list_del(&uobj->list);
mutex_unlock(&file->mutex);
kfree(uobj);
out:
mutex_unlock(&ib_uverbs_idr_mutex);
return ret ? ret : in_len;
}
/*
* Open and initialize an Interface Adapter.
* o initializes fields of struct rpcrdma_ia, including
* interface and provider attributes and protection zone.
*/
int
rpcrdma_ia_open(struct rpcrdma_xprt *xprt, struct sockaddr *addr, int memreg)
{
int rc, mem_priv;
struct rpcrdma_ia *ia = &xprt->rx_ia;
struct ib_device_attr *devattr = &ia->ri_devattr;
ia->ri_id = rpcrdma_create_id(xprt, ia, addr);
if (IS_ERR(ia->ri_id)) {
rc = PTR_ERR(ia->ri_id);
goto out1;
}
ia->ri_pd = ib_alloc_pd(ia->ri_id->device);
if (IS_ERR(ia->ri_pd)) {
rc = PTR_ERR(ia->ri_pd);
dprintk("RPC: %s: ib_alloc_pd() failed %i\n",
__func__, rc);
goto out2;
}
rc = ib_query_device(ia->ri_id->device, devattr);
if (rc) {
dprintk("RPC: %s: ib_query_device failed %d\n",
__func__, rc);
goto out3;
}
if (devattr->device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY) {
ia->ri_have_dma_lkey = 1;
ia->ri_dma_lkey = ia->ri_id->device->local_dma_lkey;
}
if (memreg == RPCRDMA_FRMR) {
/* Requires both frmr reg and local dma lkey */
if (((devattr->device_cap_flags &
(IB_DEVICE_MEM_MGT_EXTENSIONS|IB_DEVICE_LOCAL_DMA_LKEY)) !=
(IB_DEVICE_MEM_MGT_EXTENSIONS|IB_DEVICE_LOCAL_DMA_LKEY)) ||
(devattr->max_fast_reg_page_list_len == 0)) {
dprintk("RPC: %s: FRMR registration "
"not supported by HCA\n", __func__);
memreg = RPCRDMA_MTHCAFMR;
}
}
if (memreg == RPCRDMA_MTHCAFMR) {
if (!ia->ri_id->device->alloc_fmr) {
dprintk("RPC: %s: MTHCAFMR registration "
"not supported by HCA\n", __func__);
memreg = RPCRDMA_ALLPHYSICAL;
}
}
/*
* Optionally obtain an underlying physical identity mapping in
* order to do a memory window-based bind. This base registration
* is protected from remote access - that is enabled only by binding
* for the specific bytes targeted during each RPC operation, and
* revoked after the corresponding completion similar to a storage
* adapter.
*/
switch (memreg) {
case RPCRDMA_FRMR:
ia->ri_ops = &rpcrdma_frwr_memreg_ops;
break;
case RPCRDMA_ALLPHYSICAL:
ia->ri_ops = &rpcrdma_physical_memreg_ops;
mem_priv = IB_ACCESS_LOCAL_WRITE |
IB_ACCESS_REMOTE_WRITE |
IB_ACCESS_REMOTE_READ;
goto register_setup;
case RPCRDMA_MTHCAFMR:
ia->ri_ops = &rpcrdma_fmr_memreg_ops;
if (ia->ri_have_dma_lkey)
break;
mem_priv = IB_ACCESS_LOCAL_WRITE;
register_setup:
ia->ri_bind_mem = ib_get_dma_mr(ia->ri_pd, mem_priv);
if (IS_ERR(ia->ri_bind_mem)) {
printk(KERN_ALERT "%s: ib_get_dma_mr for "
"phys register failed with %lX\n",
__func__, PTR_ERR(ia->ri_bind_mem));
rc = -ENOMEM;
goto out3;
}
break;
default:
printk(KERN_ERR "RPC: Unsupported memory "
"registration mode: %d\n", memreg);
rc = -ENOMEM;
goto out3;
}
dprintk("RPC: %s: memory registration strategy is '%s'\n",
__func__, ia->ri_ops->ro_displayname);
/* Else will do memory reg/dereg for each chunk */
ia->ri_memreg_strategy = memreg;
rwlock_init(&ia->ri_qplock);
return 0;
out3:
ib_dealloc_pd(ia->ri_pd);
ia->ri_pd = NULL;
out2:
rdma_destroy_id(ia->ri_id);
ia->ri_id = NULL;
out1:
return rc;
}
static struct isert_device *isert_device_create(struct ib_device *ib_dev)
{
struct isert_device *isert_dev;
struct ib_device_attr *dev_attr;
int cqe_num, err;
struct ib_pd *pd;
struct ib_mr *mr;
struct ib_cq *cq;
char wq_name[64];
int i, j;
TRACE_ENTRY();
isert_dev = kzalloc(sizeof(*isert_dev), GFP_KERNEL);
if (unlikely(isert_dev == NULL)) {
pr_err("Failed to allocate iser dev\n");
err = -ENOMEM;
goto out;
}
dev_attr = &isert_dev->device_attr;
err = ib_query_device(ib_dev, dev_attr);
if (unlikely(err)) {
pr_err("Failed to query device, err: %d\n", err);
goto fail_query;
}
isert_dev->num_cqs = min_t(int, num_online_cpus(),
ib_dev->num_comp_vectors);
isert_dev->cq_qps = kzalloc(sizeof(*isert_dev->cq_qps) * isert_dev->num_cqs,
GFP_KERNEL);
if (unlikely(isert_dev->cq_qps == NULL)) {
pr_err("Failed to allocate iser cq_qps\n");
err = -ENOMEM;
goto fail_cq_qps;
}
isert_dev->cq_desc = vmalloc(sizeof(*isert_dev->cq_desc) * isert_dev->num_cqs);
if (unlikely(isert_dev->cq_desc == NULL)) {
pr_err("Failed to allocate %ld bytes for iser cq_desc\n",
sizeof(*isert_dev->cq_desc) * isert_dev->num_cqs);
err = -ENOMEM;
goto fail_alloc_cq_desc;
}
pd = ib_alloc_pd(ib_dev);
if (unlikely(IS_ERR(pd))) {
err = PTR_ERR(pd);
pr_err("Failed to alloc iser dev pd, err:%d\n", err);
goto fail_pd;
}
mr = ib_get_dma_mr(pd, IB_ACCESS_LOCAL_WRITE);
if (unlikely(IS_ERR(mr))) {
err = PTR_ERR(mr);
pr_err("Failed to get dma mr, err: %d\n", err);
goto fail_mr;
}
cqe_num = min(isert_dev->device_attr.max_cqe, ISER_CQ_ENTRIES);
cqe_num = cqe_num / isert_dev->num_cqs;
#ifdef CONFIG_SCST_EXTRACHECKS
if (isert_dev->device_attr.max_cqe == 0)
pr_err("Zero max_cqe encountered: you may have a compilation problem\n");
#endif
for (i = 0; i < isert_dev->num_cqs; ++i) {
struct isert_cq *cq_desc = &isert_dev->cq_desc[i];
cq_desc->dev = isert_dev;
cq_desc->idx = i;
#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 20)
INIT_WORK(&cq_desc->cq_comp_work, isert_cq_comp_work_cb, NULL);
#else
INIT_WORK(&cq_desc->cq_comp_work, isert_cq_comp_work_cb);
#endif
snprintf(wq_name, sizeof(wq_name), "isert_cq_%p", cq_desc);
#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36)
cq_desc->cq_workqueue = create_singlethread_workqueue(wq_name);
#else
#if LINUX_VERSION_CODE == KERNEL_VERSION(2, 6, 36)
cq_desc->cq_workqueue = alloc_workqueue(wq_name,
WQ_CPU_INTENSIVE|
WQ_RESCUER, 1);
#else
cq_desc->cq_workqueue = alloc_workqueue(wq_name,
WQ_CPU_INTENSIVE|
WQ_MEM_RECLAIM, 1);
#endif
#endif
if (unlikely(!cq_desc->cq_workqueue)) {
pr_err("Failed to alloc iser cq work queue for dev:%s\n",
ib_dev->name);
err = -ENOMEM;
goto fail_cq;
}
cq = ib_create_cq(ib_dev,
isert_cq_comp_handler,
isert_async_evt_handler,
cq_desc, /* context */
cqe_num,
i); /* completion vector */
if (unlikely(IS_ERR(cq))) {
cq_desc->cq = NULL;
err = PTR_ERR(cq);
pr_err("Failed to create iser dev cq, err:%d\n", err);
goto fail_cq;
}
cq_desc->cq = cq;
err = ib_req_notify_cq(cq, IB_CQ_NEXT_COMP | IB_CQ_REPORT_MISSED_EVENTS);
if (unlikely(err)) {
pr_err("Failed to request notify cq, err: %d\n", err);
goto fail_cq;
}
}
isert_dev->ib_dev = ib_dev;
isert_dev->pd = pd;
isert_dev->mr = mr;
INIT_LIST_HEAD(&isert_dev->conn_list);
lockdep_assert_held(&dev_list_mutex);
isert_dev_list_add(isert_dev);
pr_info("iser created device:%p\n", isert_dev);
return isert_dev;
fail_cq:
for (j = 0; j <= i; ++j) {
if (isert_dev->cq_desc[j].cq)
ib_destroy_cq(isert_dev->cq_desc[j].cq);
if (isert_dev->cq_desc[j].cq_workqueue)
destroy_workqueue(isert_dev->cq_desc[j].cq_workqueue);
}
ib_dereg_mr(mr);
fail_mr:
ib_dealloc_pd(pd);
fail_pd:
vfree(isert_dev->cq_desc);
fail_alloc_cq_desc:
kfree(isert_dev->cq_qps);
fail_cq_qps:
fail_query:
kfree(isert_dev);
out:
TRACE_EXIT_RES(err);
return ERR_PTR(err);
}
static int create_umr_res(struct mlx5_ib_dev *dev)
{
struct ib_qp_init_attr *init_attr = NULL;
struct ib_qp_attr *attr = NULL;
struct ib_pd *pd;
struct ib_cq *cq;
struct ib_qp *qp;
struct ib_cq_init_attr cq_attr = {};
int ret;
attr = kzalloc(sizeof(*attr), GFP_KERNEL);
init_attr = kzalloc(sizeof(*init_attr), GFP_KERNEL);
if (!attr || !init_attr) {
ret = -ENOMEM;
goto error_0;
}
pd = ib_alloc_pd(&dev->ib_dev);
if (IS_ERR(pd)) {
mlx5_ib_dbg(dev, "Couldn't create PD for sync UMR QP\n");
ret = PTR_ERR(pd);
goto error_0;
}
cq_attr.cqe = 128;
cq = ib_create_cq(&dev->ib_dev, mlx5_umr_cq_handler, NULL, NULL,
&cq_attr);
if (IS_ERR(cq)) {
mlx5_ib_dbg(dev, "Couldn't create CQ for sync UMR QP\n");
ret = PTR_ERR(cq);
goto error_2;
}
ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
init_attr->send_cq = cq;
init_attr->recv_cq = cq;
init_attr->sq_sig_type = IB_SIGNAL_ALL_WR;
init_attr->cap.max_send_wr = MAX_UMR_WR;
init_attr->cap.max_send_sge = 1;
init_attr->qp_type = MLX5_IB_QPT_REG_UMR;
init_attr->port_num = 1;
qp = mlx5_ib_create_qp(pd, init_attr, NULL);
if (IS_ERR(qp)) {
mlx5_ib_dbg(dev, "Couldn't create sync UMR QP\n");
ret = PTR_ERR(qp);
goto error_3;
}
qp->device = &dev->ib_dev;
qp->real_qp = qp;
qp->uobject = NULL;
qp->qp_type = MLX5_IB_QPT_REG_UMR;
attr->qp_state = IB_QPS_INIT;
attr->port_num = 1;
ret = mlx5_ib_modify_qp(qp, attr, IB_QP_STATE | IB_QP_PKEY_INDEX |
IB_QP_PORT, NULL);
if (ret) {
mlx5_ib_dbg(dev, "Couldn't modify UMR QP\n");
goto error_4;
}
memset(attr, 0, sizeof(*attr));
attr->qp_state = IB_QPS_RTR;
attr->path_mtu = IB_MTU_256;
ret = mlx5_ib_modify_qp(qp, attr, IB_QP_STATE, NULL);
if (ret) {
mlx5_ib_dbg(dev, "Couldn't modify umr QP to rtr\n");
goto error_4;
}
memset(attr, 0, sizeof(*attr));
attr->qp_state = IB_QPS_RTS;
ret = mlx5_ib_modify_qp(qp, attr, IB_QP_STATE, NULL);
if (ret) {
mlx5_ib_dbg(dev, "Couldn't modify umr QP to rts\n");
goto error_4;
}
dev->umrc.qp = qp;
dev->umrc.cq = cq;
dev->umrc.pd = pd;
sema_init(&dev->umrc.sem, MAX_UMR_WR);
ret = mlx5_mr_cache_init(dev);
if (ret) {
mlx5_ib_warn(dev, "mr cache init failed %d\n", ret);
goto error_4;
}
kfree(attr);
kfree(init_attr);
return 0;
error_4:
mlx5_ib_destroy_qp(qp);
error_3:
ib_destroy_cq(cq);
error_2:
ib_dealloc_pd(pd);
error_0:
kfree(attr);
kfree(init_attr);
return ret;
}
static void krping_free_qp(struct krping_cb *cb)
{
ib_destroy_qp(cb->qp);
ib_destroy_cq(cb->cq);
ib_dealloc_pd(cb->pd);
}
/*
* Open and initialize an Interface Adapter.
* o initializes fields of struct rpcrdma_ia, including
* interface and provider attributes and protection zone.
*/
int
rpcrdma_ia_open(struct rpcrdma_xprt *xprt, struct sockaddr *addr, int memreg)
{
struct rpcrdma_ia *ia = &xprt->rx_ia;
struct ib_device_attr *devattr = &ia->ri_devattr;
int rc;
ia->ri_dma_mr = NULL;
ia->ri_id = rpcrdma_create_id(xprt, ia, addr);
if (IS_ERR(ia->ri_id)) {
rc = PTR_ERR(ia->ri_id);
goto out1;
}
ia->ri_device = ia->ri_id->device;
ia->ri_pd = ib_alloc_pd(ia->ri_device);
if (IS_ERR(ia->ri_pd)) {
rc = PTR_ERR(ia->ri_pd);
dprintk("RPC: %s: ib_alloc_pd() failed %i\n",
__func__, rc);
goto out2;
}
rc = ib_query_device(ia->ri_device, devattr);
if (rc) {
dprintk("RPC: %s: ib_query_device failed %d\n",
__func__, rc);
goto out3;
}
if (memreg == RPCRDMA_FRMR) {
if (!(devattr->device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS) ||
(devattr->max_fast_reg_page_list_len == 0)) {
dprintk("RPC: %s: FRMR registration "
"not supported by HCA\n", __func__);
memreg = RPCRDMA_MTHCAFMR;
}
}
if (memreg == RPCRDMA_MTHCAFMR) {
if (!ia->ri_device->alloc_fmr) {
dprintk("RPC: %s: MTHCAFMR registration "
"not supported by HCA\n", __func__);
rc = -EINVAL;
goto out3;
}
}
switch (memreg) {
case RPCRDMA_FRMR:
ia->ri_ops = &rpcrdma_frwr_memreg_ops;
break;
case RPCRDMA_ALLPHYSICAL:
ia->ri_ops = &rpcrdma_physical_memreg_ops;
break;
case RPCRDMA_MTHCAFMR:
ia->ri_ops = &rpcrdma_fmr_memreg_ops;
break;
default:
printk(KERN_ERR "RPC: Unsupported memory "
"registration mode: %d\n", memreg);
rc = -ENOMEM;
goto out3;
}
dprintk("RPC: %s: memory registration strategy is '%s'\n",
__func__, ia->ri_ops->ro_displayname);
rwlock_init(&ia->ri_qplock);
return 0;
out3:
ib_dealloc_pd(ia->ri_pd);
ia->ri_pd = NULL;
out2:
rpcrdma_destroy_id(ia->ri_id);
ia->ri_id = NULL;
out1:
return rc;
}
/* A vanilla 2.6.19 or older kernel without backported OFED kernel headers. */
static void isert_cq_comp_work_cb(void *ctx)
{
struct isert_cq *cq_desc = ctx;
#else
static void isert_cq_comp_work_cb(struct work_struct *work)
{
struct isert_cq *cq_desc =
container_of(work, struct isert_cq, cq_comp_work);
#endif
int ret;
TRACE_ENTRY();
ret = isert_poll_cq(cq_desc);
if (unlikely(ret < 0)) { /* poll error */
pr_err("ib_poll_cq failed\n");
goto out;
}
ib_req_notify_cq(cq_desc->cq,
IB_CQ_NEXT_COMP | IB_CQ_REPORT_MISSED_EVENTS);
/*
* not all HCAs support IB_CQ_REPORT_MISSED_EVENTS,
* so we need to make sure we don't miss any events between
* last call to ib_poll_cq() and ib_req_notify_cq()
*/
isert_poll_cq(cq_desc);
out:
TRACE_EXIT();
return;
}
static void isert_cq_comp_handler(struct ib_cq *cq, void *context)
{
struct isert_cq *cq_desc = context;
#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 20)
queue_work(cq_desc->cq_workqueue, &cq_desc->cq_comp_work);
#else
queue_work_on(smp_processor_id(), cq_desc->cq_workqueue,
&cq_desc->cq_comp_work);
#endif
}
static const char *ib_event_type_str(enum ib_event_type ev_type)
{
switch (ev_type) {
case IB_EVENT_COMM_EST:
return "COMM_EST";
case IB_EVENT_QP_FATAL:
return "QP_FATAL";
case IB_EVENT_QP_REQ_ERR:
return "QP_REQ_ERR";
case IB_EVENT_QP_ACCESS_ERR:
return "QP_ACCESS_ERR";
case IB_EVENT_SQ_DRAINED:
return "SQ_DRAINED";
case IB_EVENT_PATH_MIG:
return "PATH_MIG";
case IB_EVENT_PATH_MIG_ERR:
return "PATH_MIG_ERR";
case IB_EVENT_QP_LAST_WQE_REACHED:
return "QP_LAST_WQE_REACHED";
case IB_EVENT_CQ_ERR:
return "CQ_ERR";
case IB_EVENT_SRQ_ERR:
return "SRQ_ERR";
case IB_EVENT_SRQ_LIMIT_REACHED:
return "SRQ_LIMIT_REACHED";
case IB_EVENT_PORT_ACTIVE:
return "PORT_ACTIVE";
case IB_EVENT_PORT_ERR:
return "PORT_ERR";
case IB_EVENT_LID_CHANGE:
return "LID_CHANGE";
case IB_EVENT_PKEY_CHANGE:
return "PKEY_CHANGE";
case IB_EVENT_SM_CHANGE:
return "SM_CHANGE";
case IB_EVENT_CLIENT_REREGISTER:
return "CLIENT_REREGISTER";
case IB_EVENT_DEVICE_FATAL:
return "DEVICE_FATAL";
default:
return "UNKNOWN";
}
}
static void isert_async_evt_handler(struct ib_event *async_ev, void *context)
{
struct isert_cq *cq = context;
struct isert_device *isert_dev = cq->dev;
struct ib_device *ib_dev = isert_dev->ib_dev;
char *dev_name = ib_dev->name;
enum ib_event_type ev_type = async_ev->event;
struct isert_connection *isert_conn;
TRACE_ENTRY();
switch (ev_type) {
case IB_EVENT_COMM_EST:
isert_conn = async_ev->element.qp->qp_context;
pr_info("conn:0x%p cm_id:0x%p dev:%s, QP evt: %s\n",
isert_conn, isert_conn->cm_id, dev_name,
ib_event_type_str(IB_EVENT_COMM_EST));
/* force "connection established" event */
rdma_notify(isert_conn->cm_id, IB_EVENT_COMM_EST);
break;
/* rest of QP-related events */
case IB_EVENT_QP_FATAL:
case IB_EVENT_QP_REQ_ERR:
case IB_EVENT_QP_ACCESS_ERR:
case IB_EVENT_SQ_DRAINED:
case IB_EVENT_PATH_MIG:
case IB_EVENT_PATH_MIG_ERR:
case IB_EVENT_QP_LAST_WQE_REACHED:
isert_conn = async_ev->element.qp->qp_context;
pr_err("conn:0x%p cm_id:0x%p dev:%s, QP evt: %s\n",
isert_conn, isert_conn->cm_id, dev_name,
ib_event_type_str(ev_type));
break;
/* CQ-related events */
case IB_EVENT_CQ_ERR:
pr_err("dev:%s CQ evt: %s\n", dev_name,
ib_event_type_str(ev_type));
break;
/* SRQ events */
case IB_EVENT_SRQ_ERR:
case IB_EVENT_SRQ_LIMIT_REACHED:
pr_err("dev:%s SRQ evt: %s\n", dev_name,
ib_event_type_str(ev_type));
break;
/* Port events */
case IB_EVENT_PORT_ACTIVE:
case IB_EVENT_PORT_ERR:
case IB_EVENT_LID_CHANGE:
case IB_EVENT_PKEY_CHANGE:
case IB_EVENT_SM_CHANGE:
case IB_EVENT_CLIENT_REREGISTER:
pr_err("dev:%s port:%d evt: %s\n",
dev_name, async_ev->element.port_num,
ib_event_type_str(ev_type));
break;
/* HCA events */
case IB_EVENT_DEVICE_FATAL:
pr_err("dev:%s HCA evt: %s\n", dev_name,
ib_event_type_str(ev_type));
break;
default:
pr_err("dev:%s evt: %s\n", dev_name,
ib_event_type_str(ev_type));
break;
}
TRACE_EXIT();
}
static struct isert_device *isert_device_create(struct ib_device *ib_dev)
{
struct isert_device *isert_dev;
struct ib_device_attr *dev_attr;
int cqe_num, err;
struct ib_pd *pd;
struct ib_mr *mr;
struct ib_cq *cq;
char wq_name[64];
int i, j;
TRACE_ENTRY();
isert_dev = kzalloc(sizeof(*isert_dev), GFP_KERNEL);
if (unlikely(isert_dev == NULL)) {
pr_err("Failed to allocate iser dev\n");
err = -ENOMEM;
goto out;
}
dev_attr = &isert_dev->device_attr;
err = ib_query_device(ib_dev, dev_attr);
if (unlikely(err)) {
pr_err("Failed to query device, err: %d\n", err);
goto fail_query;
}
isert_dev->num_cqs = min_t(int, num_online_cpus(),
ib_dev->num_comp_vectors);
isert_dev->cq_qps = kzalloc(sizeof(*isert_dev->cq_qps) * isert_dev->num_cqs,
GFP_KERNEL);
if (unlikely(isert_dev->cq_qps == NULL)) {
pr_err("Failed to allocate iser cq_qps\n");
err = -ENOMEM;
goto fail_cq_qps;
}
isert_dev->cq_desc = vmalloc(sizeof(*isert_dev->cq_desc) * isert_dev->num_cqs);
if (unlikely(isert_dev->cq_desc == NULL)) {
pr_err("Failed to allocate %ld bytes for iser cq_desc\n",
sizeof(*isert_dev->cq_desc) * isert_dev->num_cqs);
err = -ENOMEM;
goto fail_alloc_cq_desc;
}
pd = ib_alloc_pd(ib_dev);
if (unlikely(IS_ERR(pd))) {
err = PTR_ERR(pd);
pr_err("Failed to alloc iser dev pd, err:%d\n", err);
goto fail_pd;
}
mr = ib_get_dma_mr(pd, IB_ACCESS_LOCAL_WRITE);
if (unlikely(IS_ERR(mr))) {
err = PTR_ERR(mr);
pr_err("Failed to get dma mr, err: %d\n", err);
goto fail_mr;
}
cqe_num = min(isert_dev->device_attr.max_cqe, ISER_CQ_ENTRIES);
cqe_num = cqe_num / isert_dev->num_cqs;
#ifdef CONFIG_SCST_EXTRACHECKS
if (isert_dev->device_attr.max_cqe == 0)
pr_err("Zero max_cqe encountered: you may have a compilation problem\n");
#endif
for (i = 0; i < isert_dev->num_cqs; ++i) {
struct isert_cq *cq_desc = &isert_dev->cq_desc[i];
cq_desc->dev = isert_dev;
cq_desc->idx = i;
#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 20)
INIT_WORK(&cq_desc->cq_comp_work, isert_cq_comp_work_cb, NULL);
#else
INIT_WORK(&cq_desc->cq_comp_work, isert_cq_comp_work_cb);
#endif
snprintf(wq_name, sizeof(wq_name), "isert_cq_%p", cq_desc);
#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 36)
cq_desc->cq_workqueue = create_singlethread_workqueue(wq_name);
#else
#if LINUX_VERSION_CODE == KERNEL_VERSION(2, 6, 36)
cq_desc->cq_workqueue = alloc_workqueue(wq_name,
WQ_CPU_INTENSIVE|
WQ_RESCUER, 1);
#else
cq_desc->cq_workqueue = alloc_workqueue(wq_name,
WQ_CPU_INTENSIVE|
WQ_MEM_RECLAIM, 1);
#endif
#endif
if (unlikely(!cq_desc->cq_workqueue)) {
pr_err("Failed to alloc iser cq work queue for dev:%s\n",
ib_dev->name);
err = -ENOMEM;
goto fail_cq;
}
#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 2, 0)
cq = ib_create_cq(ib_dev,
isert_cq_comp_handler,
isert_async_evt_handler,
cq_desc, /* context */
cqe_num,
i); /* completion vector */
#else
{
struct ib_cq_init_attr ia = {
.cqe = cqe_num,
.comp_vector = i,
};
cq = ib_create_cq(ib_dev,
isert_cq_comp_handler,
isert_async_evt_handler,
cq_desc, /* context */
&ia);
}
#endif
if (unlikely(IS_ERR(cq))) {
cq_desc->cq = NULL;
err = PTR_ERR(cq);
pr_err("Failed to create iser dev cq, err:%d\n", err);
goto fail_cq;
}
cq_desc->cq = cq;
err = ib_req_notify_cq(cq, IB_CQ_NEXT_COMP | IB_CQ_REPORT_MISSED_EVENTS);
if (unlikely(err)) {
pr_err("Failed to request notify cq, err: %d\n", err);
goto fail_cq;
}
}
isert_dev->ib_dev = ib_dev;
isert_dev->pd = pd;
isert_dev->mr = mr;
INIT_LIST_HEAD(&isert_dev->conn_list);
lockdep_assert_held(&dev_list_mutex);
isert_dev_list_add(isert_dev);
pr_info("iser created device:%p\n", isert_dev);
return isert_dev;
fail_cq:
for (j = 0; j <= i; ++j) {
if (isert_dev->cq_desc[j].cq)
ib_destroy_cq(isert_dev->cq_desc[j].cq);
if (isert_dev->cq_desc[j].cq_workqueue)
destroy_workqueue(isert_dev->cq_desc[j].cq_workqueue);
}
ib_dereg_mr(mr);
fail_mr:
ib_dealloc_pd(pd);
fail_pd:
vfree(isert_dev->cq_desc);
fail_alloc_cq_desc:
kfree(isert_dev->cq_qps);
fail_cq_qps:
fail_query:
kfree(isert_dev);
out:
TRACE_EXIT_RES(err);
return ERR_PTR(err);
}
static void isert_device_release(struct isert_device *isert_dev)
{
int err, i;
TRACE_ENTRY();
lockdep_assert_held(&dev_list_mutex);
isert_dev_list_remove(isert_dev); /* remove from global list */
for (i = 0; i < isert_dev->num_cqs; ++i) {
struct isert_cq *cq_desc = &isert_dev->cq_desc[i];
#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 22)
/*
* cancel_work_sync() was introduced in 2.6.22. We can
* only wait until all scheduled work is done.
*/
flush_workqueue(cq_desc->cq_workqueue);
#else
cancel_work_sync(&cq_desc->cq_comp_work);
#endif
err = ib_destroy_cq(cq_desc->cq);
if (unlikely(err))
pr_err("Failed to destroy cq, err:%d\n", err);
destroy_workqueue(cq_desc->cq_workqueue);
}
err = ib_dereg_mr(isert_dev->mr);
if (unlikely(err))
pr_err("Failed to destroy mr, err:%d\n", err);
err = ib_dealloc_pd(isert_dev->pd);
if (unlikely(err))
pr_err("Failed to destroy pd, err:%d\n", err);
vfree(isert_dev->cq_desc);
isert_dev->cq_desc = NULL;
kfree(isert_dev->cq_qps);
isert_dev->cq_qps = NULL;
kfree(isert_dev);
TRACE_EXIT();
}
ssize_t ib_uverbs_alloc_pd(struct ib_uverbs_file *file,
const char __user *buf,
int in_len, int out_len)
{
struct ib_uverbs_alloc_pd cmd;
struct ib_uverbs_alloc_pd_resp resp;
struct ib_udata udata;
struct ib_uobject *uobj;
struct ib_pd *pd;
int ret;
if (out_len < sizeof resp)
return -ENOSPC;
if (copy_from_user(&cmd, buf, sizeof cmd))
return -EFAULT;
INIT_UDATA(&udata, buf + sizeof cmd,
(unsigned long) cmd.response + sizeof resp,
in_len - sizeof cmd, out_len - sizeof resp);
uobj = kmalloc(sizeof *uobj, GFP_KERNEL);
if (!uobj)
return -ENOMEM;
uobj->context = file->ucontext;
pd = file->device->ib_dev->alloc_pd(file->device->ib_dev,
file->ucontext, &udata);
if (IS_ERR(pd)) {
ret = PTR_ERR(pd);
goto err;
}
pd->device = file->device->ib_dev;
pd->uobject = uobj;
atomic_set(&pd->usecnt, 0);
mutex_lock(&ib_uverbs_idr_mutex);
retry:
if (!idr_pre_get(&ib_uverbs_pd_idr, GFP_KERNEL)) {
ret = -ENOMEM;
goto err_up;
}
ret = idr_get_new(&ib_uverbs_pd_idr, pd, &uobj->id);
if (ret == -EAGAIN)
goto retry;
if (ret)
goto err_up;
memset(&resp, 0, sizeof resp);
resp.pd_handle = uobj->id;
if (copy_to_user((void __user *) (unsigned long) cmd.response,
&resp, sizeof resp)) {
ret = -EFAULT;
goto err_idr;
}
mutex_lock(&file->mutex);
list_add_tail(&uobj->list, &file->ucontext->pd_list);
mutex_unlock(&file->mutex);
mutex_unlock(&ib_uverbs_idr_mutex);
return in_len;
err_idr:
idr_remove(&ib_uverbs_pd_idr, uobj->id);
err_up:
mutex_unlock(&ib_uverbs_idr_mutex);
ib_dealloc_pd(pd);
err:
kfree(uobj);
return ret;
}
