Received-SPF: pass (google.com: domain of linux-nfs-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Subject: Re: [PATCH v2 1/6] xprtrdma: Remove FMR support in
 rpcrdma_convert_iovs()
To:     Chuck Lever <chuck.lever@oracle.com>, linux-nfs@vger.kernel.org,
        linux-rdma@vger.kernel.org
References: <161236925476.1030487.10407536259816633879.stgit@manet.1015granger.net>
 <161236943446.1030487.4542967452464402073.stgit@manet.1015granger.net>
From:   Tom Talpey <tom@talpey.com>
Message-ID: <a2a7d8a1-9372-04ac-7ad6-f29c34ef3804@talpey.com>
Date:   Wed, 3 Feb 2021 13:06:49 -0500
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:68.0) Gecko/20100101
 Thunderbird/68.12.1
MIME-Version: 1.0
In-Reply-To: <161236943446.1030487.4542967452464402073.stgit@manet.1015granger.net>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 2/3/2021 11:23 AM, Chuck Lever wrote:
> Support for FMR was removed by commit ba69cd122ece ("xprtrdma:
> Remove support for FMR memory registration") [Dec 2018]. That means
> the buffer-splitting behavior of rpcrdma_convert_kvec(), added by
> commit 821c791a0bde ("xprtrdma: Segment head and tail XDR buffers
> on page boundaries") [Mar 2016], is no longer necessary. FRWR
> memory registration handles this case with aplomb.
> 
> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
> ---
>   net/sunrpc/xprtrdma/rpc_rdma.c |   19 ++++---------------
>   1 file changed, 4 insertions(+), 15 deletions(-)
> 
> diff --git a/net/sunrpc/xprtrdma/rpc_rdma.c b/net/sunrpc/xprtrdma/rpc_rdma.c
> index 8f5d0cb68360..832765f3ebba 100644
> --- a/net/sunrpc/xprtrdma/rpc_rdma.c
> +++ b/net/sunrpc/xprtrdma/rpc_rdma.c
> @@ -204,9 +204,7 @@ rpcrdma_alloc_sparse_pages(struct xdr_buf *buf)
>   	return 0;
>   }
>   
> -/* Split @vec on page boundaries into SGEs. FMR registers pages, not
> - * a byte range. Other modes coalesce these SGEs into a single MR
> - * when they can.
> +/* Convert @vec to a single SGL element.
>    *
>    * Returns pointer to next available SGE, and bumps the total number
>    * of SGEs consumed.
> @@ -215,21 +213,12 @@ static struct rpcrdma_mr_seg *
>   rpcrdma_convert_kvec(struct kvec *vec, struct rpcrdma_mr_seg *seg,
>   		     unsigned int *n)
>   {
> -	u32 remaining, page_offset;
> -	char *base;
> -
> -	base = vec->iov_base;
> -	page_offset = offset_in_page(base);
> -	remaining = vec->iov_len;
> -	while (remaining) {
> +	if (vec->iov_len) {

This is weird. Is it ever possible for a zero-length segment to be
passed? If so, it's obviously wrong to return an uninitialized "seg"
to the caller. I'd suggest simply asserting that iov_len is != 0.

I guess this was an issue in the existing code, but there, it would
only trigger if *all* the segs were zero.

>   		seg->mr_page = NULL;
> -		seg->mr_offset = base;
> -		seg->mr_len = min_t(u32, PAGE_SIZE - page_offset, remaining);
> -		remaining -= seg->mr_len;
> -		base += seg->mr_len;
> +		seg->mr_offset = vec->iov_base;

I thought the previous discussion was that this should be offset_in_page
not the (virtual) iov_base.

Tom.

> +		seg->mr_len = vec->iov_len;
>   		++seg;
>   		++(*n);
> -		page_offset = 0;
>   	}
>   	return seg;
>   }
> 
> 
>