[V2] ixgbe: Allow disabling VFs when they are pre-existing

Right now if VFs existing when the driver is loaded, it is not
possible to destroy those VFs, even though messages from the driver
suggest doing that when trying to change the number. This change
permits the disabling of SR-IOV for the case when there are
pre-existing VFs.

Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
---
Changes in V2:
- Remove unwanted content in the changelog message - sorry about that
---
 src/CORE/ixgbe_sriov.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

On Fri, 2017-03-31 at 17:26 -0700, Mark D Rustad wrote:
> Right now if VFs existing when the driver is loaded, it is not
> possible to destroy those VFs, even though messages from the driver
> suggest doing that when trying to change the number. This change
> permits the disabling of SR-IOV for the case when there are
> pre-existing VFs.
> 
> Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
> ---
> Changes in V2:
> - Remove unwanted content in the changelog message - sorry about that
> ---
>  src/CORE/ixgbe_sriov.c |    2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/src/CORE/ixgbe_sriov.c b/src/CORE/ixgbe_sriov.c
> index 9b7d05110fe9..231acf274a70 100644
> --- a/src/CORE/ixgbe_sriov.c
> +++ b/src/CORE/ixgbe_sriov.c
> @@ -603,7 +603,7 @@ static int ixgbe_pci_sriov_disable(struct pci_dev *dev)
>  	u32 current_flags = adapter->flags;
>  #endif
>  
> -	if (adapter->num_vfs == 0)
> +	if (!adapter->num_vfs && !pci_num_vf(dev))
>  		return -EINVAL;
>  
>  	err = ixgbe_disable_sriov(adapter);

So I assume this patch isn't meant for upstream at all since there
isn't a CORE folder in the upstream kernel. Also this patch does't
apply to any existing upstream tree.

On a side note it would probably be best to just drop this check
entirely. Checking for adapter->num_vfs doesn't tell you anything other
than the fact that memory is allocated for vfinfo.

In the upstream driver one thing we should probably do is add a check
against pci_num_vf() in ixgbe_disable_sriov instead of using
"!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)". That way we will try to
disable SR-IOV if that is what we are actually trying to do instead of
only if the driver had allocated memory for it.

However like I stated on the other patch it would probably be best to
take a look at fm10k and borrow from that since I think what we really
should be doing is attempting to resume with the same number of VFs
that were previously enabled so the VFs can resume operation if the PF
is reloaded with support for SR-IOV.

- Alex

Alex,

> On Apr 1, 2017, at 10:03 AM, Duyck, Alexander H <alexander.h.duyck@intel.com> wrote:
> 
> On Fri, 2017-03-31 at 17:26 -0700, Mark D Rustad wrote:
>> Right now if VFs existing when the driver is loaded, it is not
>> possible to destroy those VFs, even though messages from the driver
>> suggest doing that when trying to change the number. This change
>> permits the disabling of SR-IOV for the case when there are
>> pre-existing VFs.
>> 
>> Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
>> ---
>> Changes in V2:
>> - Remove unwanted content in the changelog message - sorry about that
>> ---
>> src/CORE/ixgbe_sriov.c |    2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>> 
>> diff --git a/src/CORE/ixgbe_sriov.c b/src/CORE/ixgbe_sriov.c
>> index 9b7d05110fe9..231acf274a70 100644
>> --- a/src/CORE/ixgbe_sriov.c
>> +++ b/src/CORE/ixgbe_sriov.c
>> @@ -603,7 +603,7 @@ static int ixgbe_pci_sriov_disable(struct pci_dev *dev)
>> 	u32 current_flags = adapter->flags;
>> #endif
>> 
>> -	if (adapter->num_vfs == 0)
>> +	if (!adapter->num_vfs && !pci_num_vf(dev))
>> 		return -EINVAL;
>> 
>> 	err = ixgbe_disable_sriov(adapter);
> 
> So I assume this patch isn't meant for upstream at all since there
> isn't a CORE folder in the upstream kernel. Also this patch does't
> apply to any existing upstream tree.

Sigh. My blunder. Sorry about the noise. I typed the command to send it in the wrong window.

> On a side note it would probably be best to just drop this check
> entirely. Checking for adapter->num_vfs doesn't tell you anything other
> than the fact that memory is allocated for vfinfo.

Well, the code here seems to be trying to avoid disabling SR-IOV when it is not enabled. Is there any need to do that other than possibly more precise messaging?

> In the upstream driver one thing we should probably do is add a check
> against pci_num_vf() in ixgbe_disable_sriov instead of using
> "!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)". That way we will try to
> disable SR-IOV if that is what we are actually trying to do instead of
> only if the driver had allocated memory for it.

I guess if the initialization path also set that bit when there were pre-existing VFs. That would provide a positive check for enablement as opposed to resource usage. Then that flag could be checked here I suppose. It would make the intent of the check here much clearer.

> However like I stated on the other patch it would probably be best to
> take a look at fm10k and borrow from that since I think what we really
> should be doing is attempting to resume with the same number of VFs
> that were previously enabled so the VFs can resume operation if the PF
> is reloaded with support for SR-IOV.

I have doubts about how feasible that is for ixgbe. I know enough to worry about that, but not enough to know that it isn't a worry. I worry that ixgbe may have too much state kept in its structures to be able to resume VFs safely.

--
Mark Rustad, Networking Division, Intel Corporation