diff mbox

[dpdk-dev,v2,2/7] net/virtio_user: postpone DRIVER OK notification

Message ID 1482477266-39199-3-git-send-email-jianfeng.tan@intel.com (mailing list archive)
State Superseded, archived
Delegated to: Yuanhan Liu
Headers

Checks

Context Check Description
ci/checkpatch success coding style OK
ci/Intel compilation success Compilation OK

Commit Message

Jianfeng Tan Dec. 23, 2016, 7:14 a.m. UTC 
  In driver probe phase, we obtain device information; and then virtio
driver will initialize device and stores info, like negotiated
features, in vhost_user layer; finally, vhost_user gets DRIVER_OK
notification from virtio driver, and then sync with backend device.

Previously, DRIVER_OK could be sent twice: 1. when ether layer invokes
eth_device_init to initialize device; 2. when user configures it with
different configuration from that of previous.

Since we can only depend on DRIVER_OK notification to sync with backend
device, we postpone it to virtio_dev_start when everything is settled.

Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
---
 drivers/net/virtio/virtio_ethdev.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

Comments

Yuanhan Liu Dec. 26, 2016, 6:27 a.m. UTC | #1 
On Fri, Dec 23, 2016 at 07:14:21AM +0000, Jianfeng Tan wrote:
> In driver probe phase, we obtain device information; and then virtio
> driver will initialize device and stores info, like negotiated
> features, in vhost_user layer; finally, vhost_user gets DRIVER_OK
> notification from virtio driver, and then sync with backend device.
> 
> Previously, DRIVER_OK could be sent twice: 1. when ether layer invokes
> eth_device_init to initialize device; 2. when user configures it with
> different configuration from that of previous.

Yes, but that's wrong. Now only 1) will be taken.

> Since we can only depend on DRIVER_OK notification to sync with backend
> device, we postpone it to virtio_dev_start when everything is settled.

I don't quite understand what you were going to fix here; you don't
state it in the commit log after all. Normally, when everything is
negotiated (when DRIVER_OK is set), we should not set it again, unless
a reset has been happened.

If you look at QEMU, qemu will simply ignore it once vhost has already
started.

	--yliu
Jianfeng Tan Dec. 26, 2016, 6:55 a.m. UTC | #2 
> -----Original Message-----
> From: Yuanhan Liu [mailto:yuanhan.liu@linux.intel.com]
> Sent: Monday, December 26, 2016 2:27 PM
> To: Tan, Jianfeng
> Cc: dev@dpdk.org; Yigit, Ferruh; Liang, Cunming
> Subject: Re: [PATCH v2 2/7] net/virtio_user: postpone DRIVER OK notification
> 
> On Fri, Dec 23, 2016 at 07:14:21AM +0000, Jianfeng Tan wrote:
> > In driver probe phase, we obtain device information; and then virtio
> > driver will initialize device and stores info, like negotiated
> > features, in vhost_user layer; finally, vhost_user gets DRIVER_OK
> > notification from virtio driver, and then sync with backend device.
> >
> > Previously, DRIVER_OK could be sent twice: 1. when ether layer invokes
> > eth_device_init to initialize device; 2. when user configures it with
> > different configuration from that of previous.
> 
> Yes, but that's wrong. Now only 1) will be taken.

From code in virtio_dev_configure() as below, if hw_ip_checksum or enable_lro is configured, the device will be reset and re-initialized.
        if (rxmode->hw_ip_checksum)
                req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
        if (rxmode->enable_lro)
                req_features |=
                        (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
                        (1ULL << VIRTIO_NET_F_GUEST_TSO6);

        /* if request features changed, reinit the device */
        if (req_features != hw->req_guest_features) {
                ret = virtio_init_device(dev, req_features);
                if (ret < 0)
                        return ret;
        }

> 
> > Since we can only depend on DRIVER_OK notification to sync with backend
> > device, we postpone it to virtio_dev_start when everything is settled.
> 
> I don't quite understand what you were going to fix here; you don't
> state it in the commit log after all. Normally, when everything is
> negotiated (when DRIVER_OK is set), we should not set it again, unless
> a reset has been happened.

I agree that DRIVER_OK should be set only if everything is settled. Under the case of hw_ip_checksum or enable_lro, a reset will be sent and the device will be re-initialized.

So my point is that since the configuration might be changed in dev_configure(), why not just send DRIVER_OK after everything is done.

> 
> If you look at QEMU, qemu will simply ignore it once vhost has already
> started.

It does not apply here. The configuration is changed, and it cannot be ignored.

Thanks,
Jianfeng

> 
> 	--yliu
Yuanhan Liu Dec. 26, 2016, 8:02 a.m. UTC | #3 
On Mon, Dec 26, 2016 at 06:55:16AM +0000, Tan, Jianfeng wrote:
> 
> 
> > -----Original Message-----
> > From: Yuanhan Liu [mailto:yuanhan.liu@linux.intel.com]
> > Sent: Monday, December 26, 2016 2:27 PM
> > To: Tan, Jianfeng
> > Cc: dev@dpdk.org; Yigit, Ferruh; Liang, Cunming
> > Subject: Re: [PATCH v2 2/7] net/virtio_user: postpone DRIVER OK notification
> > 
> > On Fri, Dec 23, 2016 at 07:14:21AM +0000, Jianfeng Tan wrote:
> > > In driver probe phase, we obtain device information; and then virtio
> > > driver will initialize device and stores info, like negotiated
> > > features, in vhost_user layer; finally, vhost_user gets DRIVER_OK
> > > notification from virtio driver, and then sync with backend device.
> > >
> > > Previously, DRIVER_OK could be sent twice: 1. when ether layer invokes
> > > eth_device_init to initialize device; 2. when user configures it with
> > > different configuration from that of previous.
> > 
> > Yes, but that's wrong. Now only 1) will be taken.
> 
> >From code in virtio_dev_configure() as below, if hw_ip_checksum or enable_lro is configured, the device will be reset and re-initialized.
>         if (rxmode->hw_ip_checksum)
>                 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
>         if (rxmode->enable_lro)
>                 req_features |=
>                         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
>                         (1ULL << VIRTIO_NET_F_GUEST_TSO6);
> 
>         /* if request features changed, reinit the device */
>         if (req_features != hw->req_guest_features) {
>                 ret = virtio_init_device(dev, req_features);
>                 if (ret < 0)
>                         return ret;
>         }

Yes, I know. But virtio_init_device() will call vtpci_reinit_complete()
at the end, doesn't it?

> > > Since we can only depend on DRIVER_OK notification to sync with backend
> > > device, we postpone it to virtio_dev_start when everything is settled.
> > 
> > I don't quite understand what you were going to fix here; you don't
> > state it in the commit log after all. Normally, when everything is
> > negotiated (when DRIVER_OK is set), we should not set it again, unless
> > a reset has been happened.
> 
> I agree that DRIVER_OK should be set only if everything is settled. Under the case of hw_ip_checksum or enable_lro, a reset will be sent and the device will be re-initialized.
> 
> So my point is that since the configuration might be changed in dev_configure(), why not just send DRIVER_OK after everything is done.
> 
> > 
> > If you look at QEMU, qemu will simply ignore it once vhost has already
> > started.
> 
> It does not apply here. The configuration is changed, and it cannot be ignored.

Why it doesn't apply here? What makes virtio_user special? If it works
to QEMU and it doesn't to virtio_user, it basically means the virtio_user
is broken ...

	--yliu
Jianfeng Tan Dec. 26, 2016, 8:26 a.m. UTC | #4 
As discussed offline, this will lead to many DRIVER OK among dev_stop()/dev_start(). I'll remove this patch.

Thanks,
Jianfeng

> -----Original Message-----
> From: Yuanhan Liu [mailto:yuanhan.liu@linux.intel.com]
> Sent: Monday, December 26, 2016 4:02 PM
> To: Tan, Jianfeng
> Cc: dev@dpdk.org; Yigit, Ferruh; Liang, Cunming
> Subject: Re: [PATCH v2 2/7] net/virtio_user: postpone DRIVER OK notification
> 
> On Mon, Dec 26, 2016 at 06:55:16AM +0000, Tan, Jianfeng wrote:
> >
> >
> > > -----Original Message-----
> > > From: Yuanhan Liu [mailto:yuanhan.liu@linux.intel.com]
> > > Sent: Monday, December 26, 2016 2:27 PM
> > > To: Tan, Jianfeng
> > > Cc: dev@dpdk.org; Yigit, Ferruh; Liang, Cunming
> > > Subject: Re: [PATCH v2 2/7] net/virtio_user: postpone DRIVER OK
> notification
> > >
> > > On Fri, Dec 23, 2016 at 07:14:21AM +0000, Jianfeng Tan wrote:
> > > > In driver probe phase, we obtain device information; and then virtio
> > > > driver will initialize device and stores info, like negotiated
> > > > features, in vhost_user layer; finally, vhost_user gets DRIVER_OK
> > > > notification from virtio driver, and then sync with backend device.
> > > >
> > > > Previously, DRIVER_OK could be sent twice: 1. when ether layer
> invokes
> > > > eth_device_init to initialize device; 2. when user configures it with
> > > > different configuration from that of previous.
> > >
> > > Yes, but that's wrong. Now only 1) will be taken.
> >
> > >From code in virtio_dev_configure() as below, if hw_ip_checksum or
> enable_lro is configured, the device will be reset and re-initialized.
> >         if (rxmode->hw_ip_checksum)
> >                 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
> >         if (rxmode->enable_lro)
> >                 req_features |=
> >                         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
> >                         (1ULL << VIRTIO_NET_F_GUEST_TSO6);
> >
> >         /* if request features changed, reinit the device */
> >         if (req_features != hw->req_guest_features) {
> >                 ret = virtio_init_device(dev, req_features);
> >                 if (ret < 0)
> >                         return ret;
> >         }
> 
> Yes, I know. But virtio_init_device() will call vtpci_reinit_complete()
> at the end, doesn't it?
> 
> > > > Since we can only depend on DRIVER_OK notification to sync with
> backend
> > > > device, we postpone it to virtio_dev_start when everything is settled.
> > >
> > > I don't quite understand what you were going to fix here; you don't
> > > state it in the commit log after all. Normally, when everything is
> > > negotiated (when DRIVER_OK is set), we should not set it again, unless
> > > a reset has been happened.
> >
> > I agree that DRIVER_OK should be set only if everything is settled. Under
> the case of hw_ip_checksum or enable_lro, a reset will be sent and the
> device will be re-initialized.
> >
> > So my point is that since the configuration might be changed in
> dev_configure(), why not just send DRIVER_OK after everything is done.
> >
> > >
> > > If you look at QEMU, qemu will simply ignore it once vhost has already
> > > started.
> >
> > It does not apply here. The configuration is changed, and it cannot be
> ignored.
> 
> Why it doesn't apply here? What makes virtio_user special? If it works
> to QEMU and it doesn't to virtio_user, it basically means the virtio_user
> is broken ...
> 
> 	--yliu
diff mbox

Patch

diff --git a/drivers/net/virtio/virtio_ethdev.c b/drivers/net/virtio/virtio_ethdev.c
index 079fd6c..d54b1bb 100644
--- a/drivers/net/virtio/virtio_ethdev.c
+++ b/drivers/net/virtio/virtio_ethdev.c
@@ -1276,7 +1276,6 @@  virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
 	ret = virtio_alloc_queues(eth_dev);
 	if (ret < 0)
 		return ret;
-	vtpci_reinit_complete(hw);
 
 	if (pci_dev)
 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
@@ -1474,6 +1473,8 @@  virtio_dev_start(struct rte_eth_dev *dev)
 	struct virtnet_tx *txvq __rte_unused;
 	struct virtio_hw *hw = dev->data->dev_private;
 
+	vtpci_reinit_complete(hw);
+
 	/* check if lsc interrupt feature is enabled */
 	if (dev->data->dev_conf.intr_conf.lsc) {
 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {