PATCH: i40e Improve trusted VF MAC addresses logging when limit is reached
Subsystems (3)
PATCH 1/2 PATCH: i40e Improve trusted VF MAC addresses logging when limit is reached
When a VF reaches the limit introduced in this commit [1], the host reports an error in the syslog but doesn't mention which VF reached its limit and what the limit is actually is which makes troubleshooting of networking
issue a bit tedious. This commit simply improves this error reporting
by adding which VF number has reached a limit and what that limit is.
[1] commit cfb1d572c986 ("i40e: Add ensurance of MacVlan resources for every trusted VF")
--- drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-)
drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c +6 -6
@@ -2932,13 +2932,14 @@ static inline int i40e_check_vf_permission(struct i40e_vf *vf, ++mac2add_cnt; } +int new_count =i40e_count_filters (vsi) + mac2add_cnt; +int max_macvlan =I40E_VC_MAX_MACVLAN_PER_TRUSTED_VF (pf->num_alloc_vfs, hw->num_ports);/* If this VF is not privileged, then we can't add more than a limited * number of addresses. Check to make sure that the additions do not * push us over the limit. */ if (!test_bit (I40E_VIRTCHNL_VF_CAP_PRIVILEGE , &vf->vf_caps)) { -if ((i40e_count_filters (vsi) + mac2add_cnt) > -I40E_VC_MAX_MAC_ADDR_PER_VF ) { +if ( new_count >I40E_VC_MAX_MAC_ADDR_PER_VF ) {dev_err (&pf->pdev->dev,"Cannot add more MAC addresses, VF is not trusted, switch the VF to trusted to add more functionality\n" );return -EPERM ;
@@ -2949,11 +2950,10 @@ static inline int i40e_check_vf_permission(struct i40e_vf *vf,* all VFs. */ }else { -if ((i40e_count_filters (vsi) + mac2add_cnt) > -I40E_VC_MAX_MACVLAN_PER_TRUSTED_VF (pf->num_alloc_vfs, - hw->num_ports)) { +if (new_count > max_macvlan) {dev_err (&pf->pdev->dev, -"Cannot add more MAC addresses, trusted VF exhausted it's resources\n" ); +"Cannot add more MAC addresses, trusted VF %d uses (%d/%d) MAC addresses\n" , + vf->vf_id, new_count, max_macvlan);return -EPERM ; } }
Signature · 7 lines
2.50.1 -- This message has been scanned for viruses and dangerous content by MailScanner, and is believed to be clean.
PATCH 2/2 PATCH: i40e Add module option to disable max VF limit
When a VF reaches the limit introduced in this commit [1], the driver refuses to add any more MACs to the filter which changes the behavior from previous releases and might break some NFVs which sometimes add more VFs than the hardcoded limit of 18 and variable limit depending
on the number of VFs created on a given PF. Disabling limit_mac_per_vf
would revert to previous behavior.
[1] commit cfb1d572c986 ("i40e: Add ensurance of MacVlan resources for every trusted VF")
--- drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)
drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c +5 -1
@@ -4,6 +4,10 @@#include "i40e.h" #include "i40e_lan_hmc.h" #include "i40e_virtchnl_pf.h" +#include <linux/moduleparam.h> + +bool __read_mostly limit_mac_per_vf =1 ; +module_param_named (limit_mac_per_vf, limit_mac_per_vf, bool,0444 );/*********************notification routines***********************/
@@ -2950,7 +2954,7 @@ static inline int i40e_check_vf_permission(struct i40e_vf *vf,* all VFs. */ }else { -if (new_count > max_macvlan) { +if (new_count > max_macvlan && limit_mac_per_vf) {dev_err (&pf->pdev->dev,"Cannot add more MAC addresses, trusted VF %d uses (%d/%d) MAC addresses\n" , vf->vf_id, new_count, max_macvlan);
Signature · 7 lines
2.50.1 -- This message has been scanned for viruses and dangerous content by MailScanner, and is believed to be clean.
Simon Horman
reviewerRe: [PATCH 2/2] PATCH: i40e Add module option to disable max VF limit
Quoted 11 lines · David Hill
When a VF reaches the limit introduced in this commit [1], the driver refuses to add any more MACs to the filter which changes the behavior from previous releases and might break some NFVs which sometimes add more VFs than the hardcoded limit of 18 and variable limit depending
on the number of VFs created on a given PF. Disablinglimit_mac_per_vf
would revert to previous behavior.[1] commit
cfb1d572c986("i40e: Add ensurance of MacVlan resources for every trusted VF")Signed-off-by David Hill <[email protected]>
Hi David,
Unfortunately adding new module parameters to Ethernet drivers is discouraged. I would suggest that devlink is an appropriate mechanism.
Signature · 1 line
pw-bot: changes-requested
Re: [PATCH 2/2] PATCH: i40e Add module option to disable max VF limit
Quoted 11 lines · Simon Horman
Quoted 10 lines · David Hill
When a VF reaches the limit introduced in this commit [1], the driver refuses to add any more MACs to the filter which changes the behavior from previous releases and might break some NFVs which sometimes add more VFs than the hardcoded limit of 18 and variable limit depending
on the number of VFs created on a given PF. Disablinglimit_mac_per_vf
would revert to previous behavior.[1] commit
cfb1d572c986("i40e: Add ensurance of MacVlan resources for every trusted VF")
Quoted 7 lines
Quoted 1 line
Signed-off-by David Hill <[email protected]>Hi David,
Unfortunately adding new module parameters to Ethernet drivers is discouraged. I would suggest that devlink is an appropriate mechanism.
At a glance, my initial suggestion would be modeling this with devlink resources?
If each VF has a devlink port in the host, we could create a resource that is the number of allowed filters for each VF, and the host could control this through the resource...
I think this might even let us nest a resource so we have one for the parent which is the total amount available and each VF port could have its amount available.
A devlink parameter could work but is a bit less flexible and doesn't show you the hierarchy with the total available filters within the PF vs what each VF is consuming.
Simon Horman
reviewerRe: [PATCH 2/2] PATCH: i40e Add module option to disable max VF limit
Quoted 29 lines · Jacob Keller
Quoted 11 lines · Simon Horman
Quoted 10 lines · David Hill
When a VF reaches the limit introduced in this commit [1], the driver refuses to add any more MACs to the filter which changes the behavior from previous releases and might break some NFVs which sometimes add more VFs than the hardcoded limit of 18 and variable limit depending
on the number of VFs created on a given PF. Disablinglimit_mac_per_vf
would revert to previous behavior.[1] commit
cfb1d572c986("i40e: Add ensurance of MacVlan resources for every trusted VF")Quoted 7 lines
Quoted 1 line
Signed-off-by David Hill <[email protected]>Hi David,
Unfortunately adding new module parameters to Ethernet drivers is discouraged. I would suggest that devlink is an appropriate mechanism.
At a glance, my initial suggestion would be modeling this with devlink resources?
If each VF has a devlink port in the host, we could create a resource that is the number of allowed filters for each VF, and the host could control this through the resource...
Quoted 7 lines
I think this might even let us nest a resource so we have one for the parent which is the total amount available and each VF port could have its amount available.
A devlink parameter could work but is a bit less flexible and doesn't show you the hierarchy with the total available filters within the PF vs what each VF is consuming.
Thanks Jacob,
For some reason I had not thought of modelling filters as a resource. But I do agree that is is a promising approach.
A message that is not in the archive (sent off-list or not yet received): CANQtZ2wffk6jUTTMYFgTYxWQBc=hmw7nAkbYB2kxt-1ihUP9Rw@mail.gmail.com
Re: [PATCH 2/2] PATCH: i40e Add module option to disable max VF limit
Hi Simon, Jacob,
I’ve also been examining this issue, as it’s affecting us.
I agree that handling the number of allowed filters per VF as a devlink resource is the best long-term approach.
However, currently in i40e, we only create a devlink port per PF and no devlink ports per VF.
Implementing the resource-per-VF approach would therefore require some extra work.
For now, could we adopt Simon’s devlink parameter suggestion as a temporary solution and consider adding the resource-based approach in the future?
Quoted 7 lines · Mohammad Heib
Hi Simon, Jacob,
I’ve also been examining this issue, as it’s affecting us.
I agree that handling the number of allowed filters per VF as a devlink resource is the best long-term approach.
However, currently in i40e, we only create a devlink port per PF and no devlink ports per VF.
Implementing the resource-per-VF approach would therefore require some extra work.
For now, could we adopt Simon’s devlink parameter suggestion as a temporary solution and consider adding the resource-based approach in the future?
Re: [PATCH 2/2] PATCH: i40e Add module option to disable max VF limit
Quoted 7 lines · mohammad heib
Hi Simon, Jacob,
I’ve also been examining this issue, as it’s affecting us.
I agree that handling the number of allowed filters per VF as a devlink resource is the best long-term approach.
However, currently in i40e, we only create a devlink port per PF and no devlink ports per VF.
Cross-PF interaction of a global resource is also tricky. Hm.
Quoted 5 lines
Implementing the resource-per-VF approach would therefore require some extra work.
For now, could we adopt Simon’s devlink parameter suggestion as a temporary solution and consider adding the resource-based approach in the future?
I do agree that its a much larger ask to implement ports and such in addition.My only concern is we would then likely want to support the parameter in perpetuity. It is generally preferred that we don't remove things in the future since it is a pain for software compatibility. Technically, I don't know if that truly falls under "user space API changing" if a single driver changes behavior.. but its definitely frowned on if done without a very good reason. Perhaps there is a way to make sure parameter works ok even once resources get added in the future, to mitigate this concern?
A message that is not in the archive (sent off-list or not yet received): [email protected]
Re: [PATCH 2/2] PATCH: i40e Add module option to disable max VF limit
Quoted 30 lines · mohammad heib
Hi
i defiantly agree with you Jacobmaybe to avoid removing the params in the future we can implement it this way:
Instead of a simple boolean parameter, we can introduce a *numeric |max_mac_per_vf| parameter*.
*
If set to |0|, the driver behaves as legacy (no per-VF limit).
*
Any non-zero value enforces that maximum number of MACs per VF.
This design addresses concerns about removing a parameter in the future because:
* The parameter can *coexist with a future devlink resource-per-VF implementation*.
* if a hierarchy of resources is added later, |max_mac_per_vf = 0| could automatically defer to the resource values, while a non-zero value can act as an override.
* This numeric parameter is more flexible than a boolean and easier to extend or adjust at runtime (via devlink in the future).
Quoted 4 lines
I think this is a reasonable *short-term solution* that preserves backward compatibility while allowing for a *long-term devlink resource-based design*.
This seems like a good approach to me.
Thanks,
Jake
Quoted 6 lines · David Hill
When a VF reaches the limit introduced in this commit [1], the host reports an error in the syslog but doesn't mention which VF reached its limit and what the limit is actually is which makes troubleshooting of networking
issue a bit tedious. This commit simply improves this error reporting
by adding which VF number has reached a limit and what that limit is.
Quoted 7 lines
[1] commit
cfb1d572c986("i40e: Add ensurance of MacVlan resources for every trusted VF")Signed-off-by David Hill <[email protected]>--- drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-)
For future reference: please observe the rule that there should be a delay of 24h between posting versions of a patch to netdev.
https://docs.kernel.org/process/maintainer-netdev.html
...