Received-SPF: pass (google.com: domain of linux-kernel+bounces-6446-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
Date: Wed, 20 Dec 2023 02:01:43 -0500
From: "Michael S. Tsirkin" <mst@redhat.com>
To: Changyuan Lyu <changyuanl@google.com>
Cc: jasowang@redhat.com, dan.j.williams@intel.com, dave.jiang@intel.com,
	linux-kernel@vger.kernel.org, nvdimm@lists.linux.dev,
	pankaj.gupta.linux@gmail.com, virtualization@lists.linux.dev,
	vishal.l.verma@intel.com, xuanzhuo@linux.alibaba.com
Subject: Re: [PATCH v2] virtio_pmem: support feature SHMEM_REGION
Message-ID: <20231220020100-mutt-send-email-mst@kernel.org>
References: <CACGkMEuEY5xJyf6H6RgqSuD0PeY9kynYywxzM2+3W6MPaav0Zw@mail.gmail.com>
 <20231220061301.228671-1-changyuanl@google.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20231220061301.228671-1-changyuanl@google.com>

On Tue, Dec 19, 2023 at 10:13:00PM -0800, Changyuan Lyu wrote:
> On Tue, Dec 19, 2023 at 7:57 PM Jason Wang <jasowang@redhat.com> wrote:
> >
> > On Tue, Dec 19, 2023 at 3:19 PM Changyuan Lyu <changyuanl@google.com> wrote:
> > >
> > > +/* shmid of the shared memory region corresponding to the pmem */
> > > +#define VIRTIO_PMEM_SHMCAP_ID 0
> >
> > NIT: not a native speaker, but any reason for "CAP" here? Would it be
> > better to use SHMMEM_REGION_ID?
> 
> I was following the name VIRTIO_FS_SHMCAP_ID_CACHE in
> include/uapi/linux/virtio_fs.h, where I guess "CAP" was referring to
> the shared memory capability when the device is on PCI bus. I agree
> SHMMEM_REGION_ID is a better name.
> 
> On Tue, Dec 19, 2023 at 3:19 PM Changyuan Lyu <changyuanl@google.com> wrote:
> >
> > +		if (!have_shm) {
> > +			dev_err(&vdev->dev, "failed to get shared memory region %d\n",
> > +					VIRTIO_PMEM_SHMCAP_ID);
> > +			return -EINVAL;
> > +		}
> 
> I realized that it needs to jump to tag out_vq to clean up vqs
> instead of simply returnning an error.
> 
> Thanks for reviewing the patch!
> 
> ---8<---
> 
> As per virtio spec 1.2 section 5.19.5.2, if the feature
> VIRTIO_PMEM_F_SHMEM_REGION has been negotiated, the driver MUST query
> shared memory ID 0 for the physical address ranges.

This is not a great description. Please describe what the patch does.

> Signed-off-by: Changyuan Lyu <changyuanl@google.com>
> 
> ---
> V2:
>   * renamed VIRTIO_PMEM_SHMCAP_ID to VIRTIO_PMEM_SHMEM_REGION_ID
>   * fixed the error handling when region 0 does not exist
> ---
>  drivers/nvdimm/virtio_pmem.c     | 30 ++++++++++++++++++++++++++----
>  include/uapi/linux/virtio_pmem.h |  8 ++++++++
>  2 files changed, 34 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/nvdimm/virtio_pmem.c b/drivers/nvdimm/virtio_pmem.c
> index a92eb172f0e7..8e447c7558cb 100644
> --- a/drivers/nvdimm/virtio_pmem.c
> +++ b/drivers/nvdimm/virtio_pmem.c
> @@ -35,6 +35,8 @@ static int virtio_pmem_probe(struct virtio_device *vdev)
>  	struct nd_region *nd_region;
>  	struct virtio_pmem *vpmem;
>  	struct resource res;
> +	struct virtio_shm_region shm_reg;
> +	bool have_shm;
>  	int err = 0;
>  
>  	if (!vdev->config->get) {
> @@ -57,10 +59,24 @@ static int virtio_pmem_probe(struct virtio_device *vdev)
>  		goto out_err;
>  	}
>  
> -	virtio_cread_le(vpmem->vdev, struct virtio_pmem_config,
> -			start, &vpmem->start);
> -	virtio_cread_le(vpmem->vdev, struct virtio_pmem_config,
> -			size, &vpmem->size);
> +	if (virtio_has_feature(vdev, VIRTIO_PMEM_F_SHMEM_REGION)) {
> +		have_shm = virtio_get_shm_region(vdev, &shm_reg,
> +				(u8)VIRTIO_PMEM_SHMEM_REGION_ID);
> +		if (!have_shm) {
> +			dev_err(&vdev->dev, "failed to get shared memory region %d\n",
> +					VIRTIO_PMEM_SHMEM_REGION_ID);
> +			err = -ENXIO;
> +			goto out_vq;
> +		}
> +		vpmem->start = shm_reg.addr;
> +		vpmem->size = shm_reg.len;
> +	} else {
> +		virtio_cread_le(vpmem->vdev, struct virtio_pmem_config,
> +				start, &vpmem->start);
> +		virtio_cread_le(vpmem->vdev, struct virtio_pmem_config,
> +				size, &vpmem->size);
> +	}
> +
>  
>  	res.start = vpmem->start;
>  	res.end   = vpmem->start + vpmem->size - 1;
> @@ -122,7 +138,13 @@ static void virtio_pmem_remove(struct virtio_device *vdev)
>  	virtio_reset_device(vdev);
>  }
>  
> +static unsigned int features[] = {
> +	VIRTIO_PMEM_F_SHMEM_REGION,
> +};
> +
>  static struct virtio_driver virtio_pmem_driver = {
> +	.feature_table		= features,
> +	.feature_table_size	= ARRAY_SIZE(features),
>  	.driver.name		= KBUILD_MODNAME,
>  	.driver.owner		= THIS_MODULE,
>  	.id_table		= id_table,
> diff --git a/include/uapi/linux/virtio_pmem.h b/include/uapi/linux/virtio_pmem.h
> index d676b3620383..c5e49b6e58b1 100644
> --- a/include/uapi/linux/virtio_pmem.h
> +++ b/include/uapi/linux/virtio_pmem.h
> @@ -14,6 +14,14 @@
>  #include <linux/virtio_ids.h>
>  #include <linux/virtio_config.h>
>  
> +/* Feature bits */
> +#define VIRTIO_PMEM_F_SHMEM_REGION 0	/* guest physical address range will be
> +					 * indicated as shared memory region 0
> +					 */
> +
> +/* shmid of the shared memory region corresponding to the pmem */
> +#define VIRTIO_PMEM_SHMEM_REGION_ID 0
> +
>  struct virtio_pmem_config {
>  	__le64 start;
>  	__le64 size;
> -- 
> 2.43.0.472.g3155946c3a-goog