summary |
shortlog |
log |
commit | commitdiff |
tree
raw |
patch |
inline | side by side (from parent 1:
252a52a)
The flushing of work requests for user QPs is implemented entirely in
the user mode library. The only kernel interaction is to mark the
user QP object indicating it is in error when the QP exits RTS. When
the user QP operations are called by the application (eg: post_send,
post_recv), the QP in error bit is checked and if set, the library
flushes the QP. If, however, the application is not doing IO, but
rather just polling the CQ, it will never get flushed work requests.
This breaks some classes of applications.
This patch adds logic to mark user CQs in error when a QP that is bound
to the CQ is marked in error. The library poll code can then notice
the CQ is in error and flush all the in error QPs bound to that CQ.
Design:
- add 1 extra CQE entry to the CQ memory that will be used to indicate
in error status.
- return the desired CQ memory size that should be mapped by the library
- bump the ABI since the create_cq uverbs response changes.
- detect older libraries and reduce the mmap size accordingly.
(The ABI bump doesn't break old libraries, since they didn't check
the ABI field anyway)
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
struct rdma_cq_setup setup;
int size = (1UL << (cq->size_log2)) * sizeof(struct t3_cqe);
struct rdma_cq_setup setup;
int size = (1UL << (cq->size_log2)) * sizeof(struct t3_cqe);
+ size += 1; /* one extra page for storing cq-in-err state */
cq->cqid = cxio_hal_get_cqid(rdev_p->rscp);
if (!cq->cqid)
return -ENOMEM;
cq->cqid = cxio_hal_get_cqid(rdev_p->rscp);
if (!cq->cqid)
return -ENOMEM;
#define CQ_VLD_ENTRY(ptr,size_log2,cqe) (Q_GENBIT(ptr,size_log2) == \
CQE_GENBIT(*cqe))
#define CQ_VLD_ENTRY(ptr,size_log2,cqe) (Q_GENBIT(ptr,size_log2) == \
CQE_GENBIT(*cqe))
+struct t3_cq_status_page {
+ u32 cq_err;
+};
+
+static inline int cxio_cq_in_error(struct t3_cq *cq)
+{
+ return ((struct t3_cq_status_page *)
+ &cq->queue[1 << cq->size_log2])->cq_err;
+}
+
+static inline void cxio_set_cq_in_error(struct t3_cq *cq)
+{
+ ((struct t3_cq_status_page *)
+ &cq->queue[1 << cq->size_log2])->cq_err = 1;
+}
+
static inline void cxio_set_wq_in_error(struct t3_wq *wq)
{
wq->queue->wq_in_err.err |= 1;
static inline void cxio_set_wq_in_error(struct t3_wq *wq)
{
wq->queue->wq_in_err.err |= 1;
atomic_inc(&qhp->refcnt);
spin_unlock(&rnicp->lock);
atomic_inc(&qhp->refcnt);
spin_unlock(&rnicp->lock);
+ if (qhp->attr.state == IWCH_QP_STATE_RTS) {
+ attrs.next_state = IWCH_QP_STATE_TERMINATE;
+ iwch_modify_qp(qhp->rhp, qhp, IWCH_QP_ATTR_NEXT_STATE,
+ &attrs, 1);
+ if (send_term)
+ iwch_post_terminate(qhp, rsp_msg);
+ }
+
event.event = ib_event;
event.device = chp->ibcq.device;
if (ib_event == IB_EVENT_CQ_ERR)
event.event = ib_event;
event.device = chp->ibcq.device;
if (ib_event == IB_EVENT_CQ_ERR)
if (qhp->ibqp.event_handler)
(*qhp->ibqp.event_handler)(&event, qhp->ibqp.qp_context);
if (qhp->ibqp.event_handler)
(*qhp->ibqp.event_handler)(&event, qhp->ibqp.qp_context);
- if (qhp->attr.state == IWCH_QP_STATE_RTS) {
- attrs.next_state = IWCH_QP_STATE_TERMINATE;
- iwch_modify_qp(qhp->rhp, qhp, IWCH_QP_ATTR_NEXT_STATE,
- &attrs, 1);
- if (send_term)
- iwch_post_terminate(qhp, rsp_msg);
- }
+ (*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context);
if (atomic_dec_and_test(&qhp->refcnt))
wake_up(&qhp->wait);
if (atomic_dec_and_test(&qhp->refcnt))
wake_up(&qhp->wait);
case TPT_ERR_BOUND:
case TPT_ERR_INVALIDATE_SHARED_MR:
case TPT_ERR_INVALIDATE_MR_WITH_MW_BOUND:
case TPT_ERR_BOUND:
case TPT_ERR_INVALIDATE_SHARED_MR:
case TPT_ERR_INVALIDATE_MR_WITH_MW_BOUND:
- (*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context);
post_qp_event(rnicp, chp, rsp_msg, IB_EVENT_QP_ACCESS_ERR, 1);
break;
post_qp_event(rnicp, chp, rsp_msg, IB_EVENT_QP_ACCESS_ERR, 1);
break;
struct iwch_create_cq_resp uresp;
struct iwch_create_cq_req ureq;
struct iwch_ucontext *ucontext = NULL;
struct iwch_create_cq_resp uresp;
struct iwch_create_cq_req ureq;
struct iwch_ucontext *ucontext = NULL;
+ static int warned;
+ size_t resplen;
PDBG("%s ib_dev %p entries %d\n", __func__, ibdev, entries);
rhp = to_iwch_dev(ibdev);
PDBG("%s ib_dev %p entries %d\n", __func__, ibdev, entries);
rhp = to_iwch_dev(ibdev);
uresp.key = ucontext->key;
ucontext->key += PAGE_SIZE;
spin_unlock(&ucontext->mmap_lock);
uresp.key = ucontext->key;
ucontext->key += PAGE_SIZE;
spin_unlock(&ucontext->mmap_lock);
- if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
+ mm->key = uresp.key;
+ mm->addr = virt_to_phys(chp->cq.queue);
+ if (udata->outlen < sizeof uresp) {
+ if (!warned++)
+ printk(KERN_WARNING MOD "Warning - "
+ "downlevel libcxgb3 (non-fatal).\n");
+ mm->len = PAGE_ALIGN((1UL << uresp.size_log2) *
+ sizeof(struct t3_cqe));
+ resplen = sizeof(struct iwch_create_cq_resp_v0);
+ } else {
+ mm->len = PAGE_ALIGN(((1UL << uresp.size_log2) + 1) *
+ sizeof(struct t3_cqe));
+ uresp.memsize = mm->len;
+ resplen = sizeof uresp;
+ }
+ if (ib_copy_to_udata(udata, &uresp, resplen)) {
kfree(mm);
iwch_destroy_cq(&chp->ibcq);
return ERR_PTR(-EFAULT);
}
kfree(mm);
iwch_destroy_cq(&chp->ibcq);
return ERR_PTR(-EFAULT);
}
- mm->key = uresp.key;
- mm->addr = virt_to_phys(chp->cq.queue);
- mm->len = PAGE_ALIGN((1UL << uresp.size_log2) *
- sizeof (struct t3_cqe));
insert_mmap(ucontext, mm);
}
PDBG("created cqid 0x%0x chp %p size 0x%0x, dma_addr 0x%0llx\n",
insert_mmap(ucontext, mm);
}
PDBG("created cqid 0x%0x chp %p size 0x%0x, dma_addr 0x%0llx\n",
dev->ibdev.post_send = iwch_post_send;
dev->ibdev.post_recv = iwch_post_receive;
dev->ibdev.get_protocol_stats = iwch_get_mib;
dev->ibdev.post_send = iwch_post_send;
dev->ibdev.post_recv = iwch_post_receive;
dev->ibdev.get_protocol_stats = iwch_get_mib;
+ dev->ibdev.uverbs_abi_ver = IWCH_UVERBS_ABI_VERSION;
dev->ibdev.iwcm = kmalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL);
if (!dev->ibdev.iwcm)
dev->ibdev.iwcm = kmalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL);
if (!dev->ibdev.iwcm)
/*
* Assumes qhp lock is held.
*/
/*
* Assumes qhp lock is held.
*/
-static void __flush_qp(struct iwch_qp *qhp, unsigned long *flag)
+static void __flush_qp(struct iwch_qp *qhp, struct iwch_cq *rchp,
+ struct iwch_cq *schp, unsigned long *flag)
- struct iwch_cq *rchp, *schp;
- rchp = get_chp(qhp->rhp, qhp->attr.rcq);
- schp = get_chp(qhp->rhp, qhp->attr.scq);
PDBG("%s qhp %p rchp %p schp %p\n", __func__, qhp, rchp, schp);
/* take a ref on the qhp since we must release the lock */
PDBG("%s qhp %p rchp %p schp %p\n", __func__, qhp, rchp, schp);
/* take a ref on the qhp since we must release the lock */
static void flush_qp(struct iwch_qp *qhp, unsigned long *flag)
{
static void flush_qp(struct iwch_qp *qhp, unsigned long *flag)
{
+ struct iwch_cq *rchp, *schp;
+
+ rchp = get_chp(qhp->rhp, qhp->attr.rcq);
+ schp = get_chp(qhp->rhp, qhp->attr.scq);
+
+ if (qhp->ibqp.uobject) {
cxio_set_wq_in_error(&qhp->wq);
cxio_set_wq_in_error(&qhp->wq);
- else
- __flush_qp(qhp, flag);
+ cxio_set_cq_in_error(&rchp->cq);
+ (*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context);
+ if (schp != rchp) {
+ cxio_set_cq_in_error(&schp->cq);
+ (*schp->ibcq.comp_handler)(&schp->ibcq,
+ schp->ibcq.cq_context);
+ }
+ return;
+ }
+ __flush_qp(qhp, rchp, schp, flag);
+struct iwch_create_cq_resp_v0 {
+ __u64 key;
+ __u32 cqid;
+ __u32 size_log2;
+};
+
struct iwch_create_cq_resp {
__u64 key;
__u32 cqid;
__u32 size_log2;
struct iwch_create_cq_resp {
__u64 key;
__u32 cqid;
__u32 size_log2;
+ __u32 memsize;
+ __u32 reserved;
};
struct iwch_create_qp_resp {
};
struct iwch_create_qp_resp {