Re: [RESEND PATCH v5 1/2] RDMA/rxe: Support RDMA Atomic Write operation

From: Li Zhijian <lizhijian@fujitsu.com>
To: "Yang, Xiao/杨 晓" <yangx.jy@fujitsu.com>,
	"linux-rdma@vger.kernel.org" <linux-rdma@vger.kernel.org>
Cc: "leon@kernel.org" <leon@kernel.org>,
	"jgg@ziepe.ca" <jgg@ziepe.ca>,
	"rpearsonhpe@gmail.com" <rpearsonhpe@gmail.com>,
	"zyjzyj2000@gmail.com" <zyjzyj2000@gmail.com>
Subject: Re: [RESEND PATCH v5 1/2] RDMA/rxe: Support RDMA Atomic Write operation
Date: Tue, 27 Sep 2022 16:18:52 +0800	[thread overview]
Message-ID: <fb02de6c-a32b-654a-62b9-55f44ffaec30@fujitsu.com> (raw)
In-Reply-To: <20220708040228.6703-2-yangx.jy@fujitsu.com>

Hi Yang

I wonder if you need to do something if a user register MR with ATOMIC_WRITE to non-rxe device, something like in my flush:

Thanks
Zhijian

  static inline int ib_check_mr_access(struct ib_device *ib_dev,
  				     unsigned int flags)
  {
+	u64 device_cap = ib_dev->attrs.device_cap_flags;
+
  	/*
  	 * Local write permission is required if remote write or
  	 * remote atomic permission is also requested.
@@ -4335,6 +4346,13 @@ static inline int ib_check_mr_access(struct ib_device *ib_dev,
  	if (flags & IB_ACCESS_ON_DEMAND &&
  	    !(ib_dev->attrs.kernel_cap_flags & IBK_ON_DEMAND_PAGING))
  		return -EINVAL;
+
+	if ((flags & IB_ACCESS_FLUSH_GLOBAL &&
+	    !(device_cap & IB_DEVICE_FLUSH_GLOBAL)) ||
+	    (flags & IB_ACCESS_FLUSH_PERSISTENT &&
+	    !(device_cap & IB_DEVICE_FLUSH_PERSISTENT)))
+		return -EINVAL;
+
  	return 0;
  }
  

On 08/07/2022 12:02, Yang, Xiao/杨 晓 wrote:
> diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
> index 9c6317cf80d5..7834285c8498 100644
> --- a/include/rdma/ib_verbs.h
> +++ b/include/rdma/ib_verbs.h

