RDMA/efa: Fix wrong resources deallocation order

When trying to destroy QP or CQ, we first decrease the refcount and
potentially free memory regions allocated for the object and then
request the device to destroy the object. If the device fails, the
object isn't fully destroyed so the user/IB core can try to destroy the
object again which will lead to underflow when trying to decrease an
already zeroed refcount.

Deallocate resources in reverse order of allocating them to safely free
them.

Fixes: ff6629f ("RDMA/efa: Do not delay freeing of DMA pages")
Reviewed-by: Michael Margolin <[email protected]>
Reviewed-by: Yossi Leybovich <[email protected]>
Signed-off-by: Yonatan Nachum <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Leon Romanovsky <[email protected]>

Author: YonatanNachum

drivers/infiniband/hw/efa/efa_verbs.c

@@ -453,12 +453,12 @@ int efa_destroy_qp(struct ib_qp *ibqp, struct ib_udata *udata)
 
 	ibdev_dbg(&dev->ibdev, "Destroy qp[%u]\n", ibqp->qp_num);
 
-	efa_qp_user_mmap_entries_remove(qp);
-
 	err = efa_destroy_qp_handle(dev, qp->qp_handle);
 	if (err)
 		return err;
 
+	efa_qp_user_mmap_entries_remove(qp);
+
 	if (qp->rq_cpu_addr) {
 		ibdev_dbg(&dev->ibdev,
 			  "qp->cpu_addr[0x%p] freed: size[%lu], dma[%pad]\n",
@@ -1017,8 +1017,8 @@ int efa_destroy_cq(struct ib_cq *ibcq, struct ib_udata *udata)
 		  "Destroy cq[%d] virt[0x%p] freed: size[%lu], dma[%pad]\n",
 		  cq->cq_idx, cq->cpu_addr, cq->size, &cq->dma_addr);
 
-	efa_cq_user_mmap_entries_remove(cq);
 	efa_destroy_cq_idx(dev, cq->cq_idx);
+	efa_cq_user_mmap_entries_remove(cq);
 	if (cq->eq) {
 		xa_erase(&dev->cqs_xa, cq->cq_idx);
 		synchronize_irq(cq->eq->irq.irqn);