On Fri, Nov 30, 2018 at 11:04:07AM +0800, oulijun wrote:
> 在 2018/11/30 8:06, Jason Gunthorpe 写道:
> > On Sat, Nov 24, 2018 at 04:49:20PM +0800, Lijun Ou wrote:
> >
> >> diff --git a/drivers/infiniband/hw/hns/hns_roce_srq.c b/drivers/infiniband/hw/hns/hns_roce_srq.c
> >> new file mode 100644
> >> index 0000000..d8a8613
> >> +++ b/drivers/infiniband/hw/hns/hns_roce_srq.c
> >> @@ -0,0 +1,26 @@
> >> +// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
> >> +/*
> >> + * Copyright (c) 2018 Hisilicon Limited.
> >> + */
> >> +
> >> +#include <rdma/ib_umem.h>
> >> +#include <rdma/hns-abi.h>
> >> +#include "hns_roce_device.h"
> >> +#include "hns_roce_cmd.h"
> >> +#include "hns_roce_hem.h"
> >> +
> >> +int hns_roce_init_srq_table(struct hns_roce_dev *hr_dev)
> >> +{
> >> + struct hns_roce_srq_table *srq_table = &hr_dev->srq_table;
> >> +
> >> + xa_init(&srq_table->xa);
> > Is a call to xa_destroy missing?
> >
> > Jason
> Hi, Jason
> It uses xarray instead radix tree. the orign code as follows:
> spin_lock_int(&srq_table->lock);
> INIT_RADIX_TREE(&srq_table->tree, GFP_ATOMIC);
>
> So, we should use xa_init instead of it. I think that it should not use xa_destroy.
Well, radix tree didn't have a destroy, it was considered destroyed
when empty, while xarray does have a destroy. Seems reasonable to call
it even if you know the xarray is empty? Matthew?
Jason
