Re: [PATCH 09/18] xen/xenbus: add xenbus_setup_ring() service function

From: Juergen Gross <jgross@suse.com>
To: Boris Ostrovsky <boris.ostrovsky@oracle.com>,
	xen-devel@lists.xenproject.org, linux-kernel@vger.kernel.org
Cc: Stefano Stabellini <sstabellini@kernel.org>
Subject: Re: [PATCH 09/18] xen/xenbus: add xenbus_setup_ring() service function
Date: Thu, 21 Apr 2022 09:57:34 +0200	[thread overview]
Message-ID: <c60ed330-a21b-24e1-e77d-c40f773c2298@suse.com> (raw)
In-Reply-To: <0d5ae87a-be43-ce7c-57b2-4567bb7e9f4d@oracle.com>

[-- Attachment #1.1.1: Type: text/plain, Size: 2082 bytes --]

On 20.04.22 20:44, Boris Ostrovsky wrote:
> 
> On 4/20/22 11:09 AM, Juergen Gross wrote:
>> +/*
>> + * xenbus_setup_ring
>> + * @dev: xenbus device
>> + * @vaddr: pointer to starting virtual address of the ring
>> + * @nr_pages: number of pages to be granted
>> + * @grefs: grant reference array to be filled in
>> + *
>> + * Allocate physically contiguous pages for a shared ring buffer and grant it
>> + * to the peer of the given device. The ring buffer is initially filled with
>> + * zeroes. The virtual address of the ring is stored at @vaddr and the
>> + * grant references are stored in the @grefs array. In case of error @vaddr
>> + * will be set to NULL and @grefs will be filled with INVALID_GRANT_REF.
>> + */
>> +int xenbus_setup_ring(struct xenbus_device *dev, gfp_t gfp, void **vaddr,
>> +              unsigned int nr_pages, grant_ref_t *grefs)
>> +{
>> +    unsigned long ring_size = nr_pages * XEN_PAGE_SIZE;
>> +    unsigned int i;
>> +    int ret;
>> +
>> +    *vaddr = alloc_pages_exact(ring_size, gfp | __GFP_ZERO);
>> +    if (!*vaddr) {
>> +        ret = -ENOMEM;
>> +        goto err;
>> +    }
>> +
>> +    ret = xenbus_grant_ring(dev, *vaddr, nr_pages, grefs);
>> +    if (ret)
>> +        goto err;
>> +
>> +    return 0;
>> +
>> + err:
>> +    if (*vaddr)
>> +        free_pages_exact(*vaddr, ring_size);
>> +    for (i = 0; i < nr_pages; i++)
>> +        grefs[i] = INVALID_GRANT_REF;
>> +    *vaddr = NULL;
>> +
>> +    return ret;
>> +}
> 
> 
> We can create a wrapper around this function that will also call 
> SHARED_RING_INIT() and FRONT_RING_INIT(). A bunch of drivers do that.

This wrapper would need to be a macro, so I decided not to do that.

It would make sense to merge the call of SHARED_RING_INIT() into
FRONT_RING_INIT() (or better, have a new macro combining the two),
though, as there is no use case of FRONT_RING_INIT() without a
call of SHARED_RING_INIT() directly before it.

Juergen

[-- Attachment #1.1.2: OpenPGP public key --]
[-- Type: application/pgp-keys, Size: 3149 bytes --]

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 495 bytes --]