Skip to contentSkip to conversation

Open-coded get_netdev implementation in MANA IB

DISCUSSION Leon Romanovsky 2 participants 5 messages Jun 17, 2024 – Jun 21, 2024 InfiniBand

lore
Share & tools
lore
#0 lore raw

Hi,

I looked again on the patch 3b73eb3a4acd ("RDMA/mana_ib: Introduce mana_ib_get_netdev helper function") and wonder if it is correct thing to do.

All new drivers shouldn't open-code get_netdev() and use the helpers ib_device_get_netdev() and ib_device_set_netdev() instead of storing netdev in the driver.

Can you please convert MANA IB driver to proper API usage?

Thanks

#1 lore raw
Quoted 10 lines

Hi,

I looked again on the patch 3b73eb3a4acd ("RDMA/mana_ib: Introduce mana_ib_get_netdev helper function") and wonder if it is correct thing to do.

All new drivers shouldn't open-code get_netdev() and use the helpers ib_device_get_netdev() and
ib_device_set_netdev() instead of storing netdev in the driver.

Can you please convert MANA IB driver to proper API usage?

Hi Leon,

This helper was introduced to remove a boiler plate code which was getting the netdev from gc->mana.driver_data. I sure can fix it and use ib_device_get_netdev api.

Should I send it to the rdma-next on the top of the latest commit?

Thanks

Quoted 2 lines

Thanks

#2 lore raw
Quoted 9 lines · Konstantin Taranov
Quoted 9 lines

Hi,

I looked again on the patch 3b73eb3a4acd ("RDMA/mana_ib: Introduce mana_ib_get_netdev helper function") and wonder if it is correct thing to do.

All new drivers shouldn't open-code get_netdev() and use the helpers ib_device_get_netdev() and
ib_device_set_netdev() instead of storing netdev in the driver.

Quoted 6 lines
Quoted 1 line

Can you please convert MANA IB driver to proper API usage?

Hi Leon,

This helper was introduced to remove a boiler plate code which was getting the netdev from gc->mana.driver_data.

Yes, the thing is that probably we were supposed to move to ib_device_get_netdev() and don't invest time in mana_ib_get_netdev() helper at all.

Quoted 3 lines

I sure can fix it and use ib_device_get_netdev api.

Should I send it to the rdma-next on the top of the latest commit?

Yes, please.

Quoted 5 lines

Thanks

Quoted 2 lines

Thanks

#3 lore raw
Quoted 10 lines

Yes, the thing is that probably we were supposed to move to ib_device_get_netdev() and don't invest time in mana_ib_get_netdev() helper at all.

Quoted 3 lines

I sure can fix it and use ib_device_get_netdev api.

Should I send it to the rdma-next on the top of the latest commit?

Yes, please.

Hi Leon,

While preparing the patch, I realized that ib_device_get_netdev is not public and it is defined in core_priv.h. Should I move the declaration and export the symbol?

Thanks

#4 lore raw
Quoted 8 lines · Konstantin Taranov
Quoted 8 lines

Yes, the thing is that probably we were supposed to move to ib_device_get_netdev() and don't invest time in mana_ib_get_netdev() helper at all.

Quoted 3 lines

I sure can fix it and use ib_device_get_netdev api.

Should I send it to the rdma-next on the top of the latest commit?

Quoted 7 lines
Quoted 2 lines

Yes, please.

Hi Leon,

While preparing the patch, I realized that ib_device_get_netdev is not public and it is defined in core_priv.h. Should I move the declaration and export the symbol?

Let's keep it as is for now. I will try to return to this task after we find a way to remove .get_netdev() callback from the drivers.

Sorry for wasting your time.

Thanks

Quoted 2 lines

Thanks