ixgbe: Allow disabling VFs when they are pre-existing
PATCH v2 1/1 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.
---
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(-)
src/CORE/ixgbe_sriov.c +1 -1
@@ -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);
Duyck
PATCH v2 1/1 ixgbe: Allow disabling VFs when they are pre-existing
Quoted 19 lines · Mark D Rustad
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 <[email protected]>---
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(-)src/CORE/ixgbe_sriov.c +0
-0
src/CORE/ixgbe_sriov.c static int ixgbe_pci_sriov_disable(struct pci_dev *dev) · line 609
Quoted 8 lines
src/CORE/ixgbe_sriov.c +1
-1@@ -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
Rustad
PATCH v2 1/1 ixgbe: Allow disabling VFs when they are pre-existing
Alex,
Quoted 29 lines · Alexander H
Quoted 26 lines · Mark D Rustad
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 <[email protected]>---
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(-)src/CORE/ixgbe_sriov.c +1
-1@@ -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 ;
Quoted 5 lines
src/CORE/ixgbe_sriov.c static int ixgbe_pci_sriov_disable(struct pci_dev *dev) · line 609
Quoted 1 line
src/CORE/ixgbe_sriov.c +0
-0err =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.
Quoted 3 lines
On a side note it would probably be best to just drop this check entirely. Checking for
adapter->num_vfsdoesn'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?
Quoted 5 lines
In the upstream driver one thing we should probably do is add a check against
pci_num_vf()inixgbe_disable_sriovinstead 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.
Quoted 5 lines
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
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 841 bytes
Desc: Message signed with OpenPGP
URL: <http://lists.osuosl.org/pipermail/intel-wired-lan/attachments/20170403/fe987d0a/attachment.asc>