Re: [iwl-next v1 6/7] ice: enable_rdma devlink param

[Jiri Pirko <jiri@resnulli.us>]

Tue, Feb 13, 2024 at 11:19:49AM CET, wojciech.drewek@intel.com wrote:
>
>
>On 13.02.2024 10:58, Michal Swiatkowski wrote:
>> On Tue, Feb 13, 2024 at 10:07:17AM +0100, Jiri Pirko wrote:
>>> Tue, Feb 13, 2024 at 08:35:08AM CET, michal.swiatkowski@linux.intel.com wrote:
>>>> Implement enable_rdma devlink parameter to allow user to turn RDMA
>>>> feature on and off.
>>>>
>>>> It is useful when there is no enough interrupts and user doesn't need
>>>> RDMA feature.
>>>>
>>>> Reviewed-by: Jan Sokolowski <jan.sokolowski@intel.com>
>>>> Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
>>>> Signed-off-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
>>>> ---
>>>> drivers/net/ethernet/intel/ice/ice_devlink.c | 32 ++++++++++++++++++--
>>>> drivers/net/ethernet/intel/ice/ice_lib.c     |  8 ++++-
>>>> drivers/net/ethernet/intel/ice/ice_main.c    | 18 +++++------
>>>> 3 files changed, 45 insertions(+), 13 deletions(-)
>>>>
>>>> diff --git a/drivers/net/ethernet/intel/ice/ice_devlink.c b/drivers/net/ethernet/intel/ice/ice_devlink.c
>>>> index b82ff9556a4b..4f048268db72 100644
>>>> --- a/drivers/net/ethernet/intel/ice/ice_devlink.c
>>>> +++ b/drivers/net/ethernet/intel/ice/ice_devlink.c
>>>> @@ -1675,6 +1675,19 @@ ice_devlink_msix_min_pf_validate(struct devlink *devlink, u32 id,
>>>> 	return 0;
>>>> }
>>>>
>>>> +static int ice_devlink_enable_rdma_validate(struct devlink *devlink, u32 id,
>>>> +					    union devlink_param_value val,
>>>> +					    struct netlink_ext_ack *extack)
>>>> +{
>>>> +	struct ice_pf *pf = devlink_priv(devlink);
>>>> +	bool new_state = val.vbool;
>>>> +
>>>> +	if (new_state && !test_bit(ICE_FLAG_RDMA_ENA, pf->flags))
>>>> +		return -EOPNOTSUPP;
>>>> +
>>>> +	return 0;
>>>> +}
>>>> +
>>>> static const struct devlink_param ice_devlink_params[] = {
>>>> 	DEVLINK_PARAM_GENERIC(ENABLE_ROCE, BIT(DEVLINK_PARAM_CMODE_RUNTIME),
>>>> 			      ice_devlink_enable_roce_get,
>>>> @@ -1700,6 +1713,8 @@ static const struct devlink_param ice_devlink_params[] = {
>>>> 			      ice_devlink_msix_min_pf_get,
>>>> 			      ice_devlink_msix_min_pf_set,
>>>> 			      ice_devlink_msix_min_pf_validate),
>>>> +	DEVLINK_PARAM_GENERIC(ENABLE_RDMA, BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
>>>> +			      NULL, NULL, ice_devlink_enable_rdma_validate),
>>>> };
>>>>
>>>> static void ice_devlink_free(void *devlink_ptr)
>>>> @@ -1776,9 +1791,22 @@ ice_devlink_set_switch_id(struct ice_pf *pf, struct netdev_phys_item_id *ppid)
>>>> int ice_devlink_register_params(struct ice_pf *pf)
>>>> {
>>>> 	struct devlink *devlink = priv_to_devlink(pf);
>>>> +	union devlink_param_value value;
>>>> +	int err;
>>>> +
>>>> +	err = devlink_params_register(devlink, ice_devlink_params,
>>>
>>> Interesting, can't you just take the lock before this and call
>>> devl_params_register()?
>>>
>> I mess up a locking here and also in subfunction patchset. I will follow
>> you suggestion and take lock for whole init/cleanup. Thanks.
>> 
>>> Moreover, could you please fix your init/cleanup paths for hold devlink
>>> instance lock the whole time?
>>>
>> You right, I will prepare patch for it.
>
>I think my patch implements your suggestion Jiri:
>https://lore.kernel.org/netdev/20240212211202.1051990-5-anthony.l.nguyen@intel.com/

This patch is based on the one in your link apparently. I think it is
incomplete. Idk


>
>> 
>>>
>>> pw-bot: cr
>>>
>>>
>>>> +				      ARRAY_SIZE(ice_devlink_params));
>>>> +	if (err)
>>>> +		return err;
>>>>
>>>> -	return devlink_params_register(devlink, ice_devlink_params,
>>>> -				       ARRAY_SIZE(ice_devlink_params));
>>>> +	devl_lock(devlink);
>>>> +	value.vbool = test_bit(ICE_FLAG_RDMA_ENA, pf->flags);
>>>> +	devl_param_driverinit_value_set(devlink,
>>>> +					DEVLINK_PARAM_GENERIC_ID_ENABLE_RDMA,
>>>> +					value);
>>>> +	devl_unlock(devlink);
>>>> +
>>>> +	return 0;
>>>> }
>>>>
>>>> void ice_devlink_unregister_params(struct ice_pf *pf)
>>>> diff --git a/drivers/net/ethernet/intel/ice/ice_lib.c b/drivers/net/ethernet/intel/ice/ice_lib.c
>>>> index a30d2d2b51c1..4d5c3d699044 100644
>>>> --- a/drivers/net/ethernet/intel/ice/ice_lib.c
>>>> +++ b/drivers/net/ethernet/intel/ice/ice_lib.c
>>>> @@ -829,7 +829,13 @@ bool ice_is_safe_mode(struct ice_pf *pf)
>>>>  */
>>>> bool ice_is_rdma_ena(struct ice_pf *pf)
>>>> {
>>>> -	return test_bit(ICE_FLAG_RDMA_ENA, pf->flags);
>>>> +	union devlink_param_value value;
>>>> +	int err;
>>>> +
>>>> +	err = devl_param_driverinit_value_get(priv_to_devlink(pf),
>>>> +					      DEVLINK_PARAM_GENERIC_ID_ENABLE_RDMA,
>>>> +					      &value);
>>>> +	return err ? false : value.vbool;
>>>> }
>>>>
>>>> /**
>>>> diff --git a/drivers/net/ethernet/intel/ice/ice_main.c b/drivers/net/ethernet/intel/ice/ice_main.c
>>>> index 824732f16112..4dd59d888ec7 100644
>>>> --- a/drivers/net/ethernet/intel/ice/ice_main.c
>>>> +++ b/drivers/net/ethernet/intel/ice/ice_main.c
>>>> @@ -5177,23 +5177,21 @@ ice_probe(struct pci_dev *pdev, const struct pci_device_id __always_unused *ent)
>>>> 	if (err)
>>>> 		goto err_init;
>>>>
>>>> +	err = ice_init_devlink(pf);
>>>> +	if (err)
>>>> +		goto err_init_devlink;
>>>> +
>>>> 	devl_lock(priv_to_devlink(pf));
>>>> 	err = ice_load(pf);
>>>> 	devl_unlock(priv_to_devlink(pf));
>>>> 	if (err)
>>>> 		goto err_load;
>>>>
>>>> -	err = ice_init_devlink(pf);
>>>> -	if (err)
>>>> -		goto err_init_devlink;
>>>> -
>>>> 	return 0;
>>>>
>>>> -err_init_devlink:
>>>> -	devl_lock(priv_to_devlink(pf));
>>>> -	ice_unload(pf);
>>>> -	devl_unlock(priv_to_devlink(pf));
>>>> err_load:
>>>> +	ice_deinit_devlink(pf);
>>>> +err_init_devlink:
>>>> 	ice_deinit(pf);
>>>> err_init:
>>>> 	pci_disable_device(pdev);
>>>> @@ -5290,12 +5288,12 @@ static void ice_remove(struct pci_dev *pdev)
>>>> 	if (!ice_is_safe_mode(pf))
>>>> 		ice_remove_arfs(pf);
>>>>
>>>> -	ice_deinit_devlink(pf);
>>>> -
>>>> 	devl_lock(priv_to_devlink(pf));
>>>> 	ice_unload(pf);
>>>> 	devl_unlock(priv_to_devlink(pf));
>>>>
>>>> +	ice_deinit_devlink(pf);
>>>> +
>>>> 	ice_deinit(pf);
>>>> 	ice_vsi_release_all(pf);
>>>>
>>>> -- 
>>>> 2.42.0
>>>>
>>>>