From: Gautam Dawar <gdawar@amd.com>
To: Jason Wang <jasowang@redhat.com>, Gautam Dawar <gautam.dawar@amd.com>
Cc: linux-net-drivers@amd.com, netdev@vger.kernel.org,
eperezma@redhat.com, tanuj.kamde@amd.com, Koushik.Dutta@amd.com,
harpreet.anand@amd.com, Edward Cree <ecree.xilinx@gmail.com>,
Martin Habets <habetsm.xilinx@gmail.com>,
"David S. Miller" <davem@davemloft.net>,
Eric Dumazet <edumazet@google.com>,
Jakub Kicinski <kuba@kernel.org>, Paolo Abeni <pabeni@redhat.com>,
linux-kernel@vger.kernel.org
Subject: Re: [PATCH net-next 08/11] sfc: implement device status related vdpa config operations
Date: Mon, 9 Jan 2023 15:51:11 +0530 [thread overview]
Message-ID: <c5956679-82c1-336b-3190-de32db1c0926@amd.com> (raw)
In-Reply-To: <CACGkMEtGCbUBZRFh7EUJyymuWZ9uxiAOeJHA6h-dGa9Y3pDZGw@mail.gmail.com>
On 12/14/22 12:15, Jason Wang wrote:
> Caution: This message originated from an External Source. Use proper caution when opening attachments, clicking links, or responding.
>
>
> On Wed, Dec 7, 2022 at 10:57 PM Gautam Dawar <gautam.dawar@amd.com> wrote:
>> vDPA config opertions to handle get/set device status and device
>> reset have been implemented.
>>
>> Signed-off-by: Gautam Dawar <gautam.dawar@amd.com>
>> ---
>> drivers/net/ethernet/sfc/ef100_vdpa.c | 7 +-
>> drivers/net/ethernet/sfc/ef100_vdpa.h | 1 +
>> drivers/net/ethernet/sfc/ef100_vdpa_ops.c | 133 ++++++++++++++++++++++
>> 3 files changed, 140 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/net/ethernet/sfc/ef100_vdpa.c b/drivers/net/ethernet/sfc/ef100_vdpa.c
>> index 04d64bfe3c93..80bca281a748 100644
>> --- a/drivers/net/ethernet/sfc/ef100_vdpa.c
>> +++ b/drivers/net/ethernet/sfc/ef100_vdpa.c
>> @@ -225,9 +225,14 @@ static int vdpa_allocate_vis(struct efx_nic *efx, unsigned int *allocated_vis)
>>
>> static void ef100_vdpa_delete(struct efx_nic *efx)
>> {
>> + struct vdpa_device *vdpa_dev;
>> +
>> if (efx->vdpa_nic) {
>> + vdpa_dev = &efx->vdpa_nic->vdpa_dev;
>> + ef100_vdpa_reset(vdpa_dev);
> Any reason we need to reset during delete?
ef100_reset_vdpa_device() does the necessary clean-up including freeing
irqs, deleting filters and deleting the vrings which is required while
removing the vdpa device or unloading the driver.
>
>> +
>> /* replace with _vdpa_unregister_device later */
>> - put_device(&efx->vdpa_nic->vdpa_dev.dev);
>> + put_device(&vdpa_dev->dev);
>> efx->vdpa_nic = NULL;
>> }
>> efx_mcdi_free_vis(efx);
>> diff --git a/drivers/net/ethernet/sfc/ef100_vdpa.h b/drivers/net/ethernet/sfc/ef100_vdpa.h
>> index a33edd6dda12..1b0bbba88154 100644
>> --- a/drivers/net/ethernet/sfc/ef100_vdpa.h
>> +++ b/drivers/net/ethernet/sfc/ef100_vdpa.h
>> @@ -186,6 +186,7 @@ int ef100_vdpa_add_filter(struct ef100_vdpa_nic *vdpa_nic,
>> enum ef100_vdpa_mac_filter_type type);
>> int ef100_vdpa_irq_vectors_alloc(struct pci_dev *pci_dev, u16 nvqs);
>> void ef100_vdpa_irq_vectors_free(void *data);
>> +int ef100_vdpa_reset(struct vdpa_device *vdev);
>>
>> static inline bool efx_vdpa_is_little_endian(struct ef100_vdpa_nic *vdpa_nic)
>> {
>> diff --git a/drivers/net/ethernet/sfc/ef100_vdpa_ops.c b/drivers/net/ethernet/sfc/ef100_vdpa_ops.c
>> index 132ddb4a647b..718b67f6da90 100644
>> --- a/drivers/net/ethernet/sfc/ef100_vdpa_ops.c
>> +++ b/drivers/net/ethernet/sfc/ef100_vdpa_ops.c
>> @@ -251,6 +251,62 @@ static bool is_qid_invalid(struct ef100_vdpa_nic *vdpa_nic, u16 idx,
>> return false;
>> }
>>
>> +static void ef100_reset_vdpa_device(struct ef100_vdpa_nic *vdpa_nic)
>> +{
>> + int i;
>> +
>> + WARN_ON(!mutex_is_locked(&vdpa_nic->lock));
>> +
>> + if (!vdpa_nic->status)
>> + return;
>> +
>> + vdpa_nic->vdpa_state = EF100_VDPA_STATE_INITIALIZED;
>> + vdpa_nic->status = 0;
>> + vdpa_nic->features = 0;
>> + for (i = 0; i < (vdpa_nic->max_queue_pairs * 2); i++)
>> + reset_vring(vdpa_nic, i);
>> +}
>> +
>> +/* May be called under the rtnl lock */
>> +int ef100_vdpa_reset(struct vdpa_device *vdev)
>> +{
>> + struct ef100_vdpa_nic *vdpa_nic = get_vdpa_nic(vdev);
>> +
>> + /* vdpa device can be deleted anytime but the bar_config
>> + * could still be vdpa and hence efx->state would be STATE_VDPA.
>> + * Accordingly, ensure vdpa device exists before reset handling
>> + */
>> + if (!vdpa_nic)
>> + return -ENODEV;
>> +
>> + mutex_lock(&vdpa_nic->lock);
>> + ef100_reset_vdpa_device(vdpa_nic);
>> + mutex_unlock(&vdpa_nic->lock);
>> + return 0;
>> +}
>> +
>> +static int start_vdpa_device(struct ef100_vdpa_nic *vdpa_nic)
>> +{
>> + int rc = 0;
>> + int i, j;
>> +
>> + for (i = 0; i < (vdpa_nic->max_queue_pairs * 2); i++) {
>> + if (can_create_vring(vdpa_nic, i)) {
>> + rc = create_vring(vdpa_nic, i);
> So I think we can safely remove the create_vring() in set_vq_ready()
> since it's undefined behaviour if set_vq_ready() is called after
> DRIVER_OK.
Is this (undefined) behavior documented in the virtio spec? If so, can
you please point me to the section of virtio spec that calls this order
(set_vq_ready() after setting DRIVER_OK) undefined? Is it just that the
queue can't be enabled after DRIVER_OK or the reverse (disabling the
queue) after DRIVER_OK is not allowed?
>
>> + if (rc)
>> + goto clear_vring;
>> + }
>> + }
>> + vdpa_nic->vdpa_state = EF100_VDPA_STATE_STARTED;
>> + return rc;
>> +
>> +clear_vring:
>> + for (j = 0; j < i; j++)
>> + if (vdpa_nic->vring[j].vring_created)
>> + delete_vring(vdpa_nic, j);
>> + return rc;
>> +}
>> +
>> static int ef100_vdpa_set_vq_address(struct vdpa_device *vdev,
>> u16 idx, u64 desc_area, u64 driver_area,
>> u64 device_area)
>> @@ -568,6 +624,80 @@ static u32 ef100_vdpa_get_vendor_id(struct vdpa_device *vdev)
>> return EF100_VDPA_VENDOR_ID;
>> }
>>
>> +static u8 ef100_vdpa_get_status(struct vdpa_device *vdev)
>> +{
>> + struct ef100_vdpa_nic *vdpa_nic = get_vdpa_nic(vdev);
>> + u8 status;
>> +
>> + mutex_lock(&vdpa_nic->lock);
>> + status = vdpa_nic->status;
>> + mutex_unlock(&vdpa_nic->lock);
>> + return status;
>> +}
>> +
>> +static void ef100_vdpa_set_status(struct vdpa_device *vdev, u8 status)
>> +{
>> + struct ef100_vdpa_nic *vdpa_nic = get_vdpa_nic(vdev);
>> + u8 new_status;
>> + int rc;
>> +
>> + mutex_lock(&vdpa_nic->lock);
>> + if (!status) {
>> + dev_info(&vdev->dev,
>> + "%s: Status received is 0. Device reset being done\n",
>> + __func__);
>> + ef100_reset_vdpa_device(vdpa_nic);
>> + goto unlock_return;
>> + }
>> + new_status = status & ~vdpa_nic->status;
>> + if (new_status == 0) {
>> + dev_info(&vdev->dev,
>> + "%s: New status same as current status\n", __func__);
>> + goto unlock_return;
>> + }
>> + if (new_status & VIRTIO_CONFIG_S_FAILED) {
>> + ef100_reset_vdpa_device(vdpa_nic);
>> + goto unlock_return;
>> + }
>> +
>> + if (new_status & VIRTIO_CONFIG_S_ACKNOWLEDGE &&
>> + vdpa_nic->vdpa_state == EF100_VDPA_STATE_INITIALIZED) {
> As replied before, I think there's no need to check
> EF100_VDPA_STATE_INITIALIZED, otherwise it could be a bug somewhere.
Ok. Will remove the check against EF100_VDPA_STATE_INITIALIZED.
>> + vdpa_nic->status |= VIRTIO_CONFIG_S_ACKNOWLEDGE;
>> + new_status &= ~VIRTIO_CONFIG_S_ACKNOWLEDGE;
>> + }
>> + if (new_status & VIRTIO_CONFIG_S_DRIVER &&
>> + vdpa_nic->vdpa_state == EF100_VDPA_STATE_INITIALIZED) {
>> + vdpa_nic->status |= VIRTIO_CONFIG_S_DRIVER;
>> + new_status &= ~VIRTIO_CONFIG_S_DRIVER;
>> + }
>> + if (new_status & VIRTIO_CONFIG_S_FEATURES_OK &&
>> + vdpa_nic->vdpa_state == EF100_VDPA_STATE_INITIALIZED) {
>> + vdpa_nic->status |= VIRTIO_CONFIG_S_FEATURES_OK;
>> + vdpa_nic->vdpa_state = EF100_VDPA_STATE_NEGOTIATED;
> I think we can simply map EF100_VDPA_STATE_NEGOTIATED to
> VIRTIO_CONFIG_S_FEATURES_OK.
>
> E.g the code doesn't fail the feature negotiation by clearing the
> VIRTIO_CONFIG_S_FEATURES_OK when ef100_vdpa_set_driver_feature fails?
Ok.
>
> Thanks
Regards,
Gautam
next prev parent reply other threads:[~2023-01-09 10:24 UTC|newest]
Thread overview: 39+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-12-07 14:54 [PATCH net-next 00/11] sfc: add vDPA support for EF100 devices Gautam Dawar
2022-12-07 14:54 ` [PATCH net-next 01/11] sfc: add function personality " Gautam Dawar
2022-12-07 14:54 ` [PATCH net-next 02/11] sfc: implement MCDI interface for vDPA operations Gautam Dawar
2022-12-14 6:43 ` Jason Wang
2022-12-16 12:50 ` Gautam Dawar
2022-12-07 14:54 ` [PATCH net-next 03/11] sfc: implement init and fini functions for vDPA personality Gautam Dawar
2022-12-07 14:54 ` [PATCH net-next 04/11] sfc: implement vDPA management device operations Gautam Dawar
2022-12-07 16:31 ` kernel test robot
2022-12-14 6:43 ` Jason Wang
2022-12-15 7:07 ` Gautam Dawar
2022-12-07 14:54 ` [PATCH net-next 05/11] sfc: implement vdpa device config operations Gautam Dawar
2022-12-14 6:44 ` Jason Wang
2022-12-15 9:53 ` Gautam Dawar
2022-12-07 14:54 ` [PATCH net-next 06/11] sfc: implement vdpa vring " Gautam Dawar
2022-12-14 6:45 ` Jason Wang
2022-12-07 14:54 ` [PATCH net-next 07/11] sfc: implement filters for receiving traffic Gautam Dawar
2022-12-14 6:45 ` Jason Wang
2023-01-05 12:46 ` Gautam Dawar
2022-12-07 14:54 ` [PATCH net-next 08/11] sfc: implement device status related vdpa config operations Gautam Dawar
2022-12-14 6:45 ` Jason Wang
2023-01-09 10:21 ` Gautam Dawar [this message]
2023-01-11 6:36 ` Jason Wang
2023-01-13 4:28 ` Jason Wang
2023-01-13 6:10 ` Gautam Dawar
2023-01-13 6:20 ` Jason Wang
2023-01-13 6:33 ` Gautam Dawar
2023-01-16 2:55 ` Jason Wang
2022-12-07 14:54 ` [PATCH net-next 09/11] sfc: implement iova rbtree to store dma mappings Gautam Dawar
2022-12-14 6:46 ` Jason Wang
2022-12-16 12:48 ` Gautam Dawar
2022-12-19 6:03 ` Jason Wang
2023-01-05 13:08 ` Gautam Dawar
2023-01-09 8:49 ` Jason Wang
2023-02-09 13:39 ` Gautam Dawar
2022-12-07 14:54 ` [PATCH net-next 10/11] sfc: implement vdpa config_ops for dma operations Gautam Dawar
2022-12-07 19:23 ` kernel test robot
2022-12-14 6:46 ` Jason Wang
2022-12-07 14:54 ` [PATCH net-next 11/11] sfc: register the vDPA device Gautam Dawar
2022-12-11 18:05 ` [PATCH net-next 00/11] sfc: add vDPA support for EF100 devices Martin Habets
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=c5956679-82c1-336b-3190-de32db1c0926@amd.com \
--to=gdawar@amd.com \
--cc=Koushik.Dutta@amd.com \
--cc=davem@davemloft.net \
--cc=ecree.xilinx@gmail.com \
--cc=edumazet@google.com \
--cc=eperezma@redhat.com \
--cc=gautam.dawar@amd.com \
--cc=habetsm.xilinx@gmail.com \
--cc=harpreet.anand@amd.com \
--cc=jasowang@redhat.com \
--cc=kuba@kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-net-drivers@amd.com \
--cc=netdev@vger.kernel.org \
--cc=pabeni@redhat.com \
--cc=tanuj.kamde@amd.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).