Re: [PATCH iwl-next] ice: store VF's pci_dev ptr in ice_vf

[Leon Romanovsky <leon@kernel.org>]

On Tue, Aug 22, 2023 at 08:46:46PM +0000, Keller, Jacob E wrote:
> 
> 
> > -----Original Message-----
> > From: Leon Romanovsky <leon@kernel.org>
> > Sent: Monday, August 21, 2023 4:02 AM
> > To: Kitszel, Przemyslaw <przemyslaw.kitszel@intel.com>
> > Cc: Nguyen, Anthony L <anthony.l.nguyen@intel.com>; Brandeburg, Jesse
> > <jesse.brandeburg@intel.com>; intel-wired-lan@lists.osuosl.org;
> > netdev@vger.kernel.org; Polchlopek, Mateusz <mateusz.polchlopek@intel.com>;
> > Keller, Jacob E <jacob.e.keller@intel.com>
> > Subject: Re: [PATCH iwl-next] ice: store VF's pci_dev ptr in ice_vf
> > 
> > On Mon, Aug 21, 2023 at 12:48:40PM +0200, Przemek Kitszel wrote:
> > > On 8/18/23 20:20, Leon Romanovsky wrote:
> > > > On Fri, Aug 18, 2023 at 02:20:51PM +0200, Przemek Kitszel wrote:
> > > > > On 8/16/23 16:31, Leon Romanovsky wrote:
> > > > > > On Wed, Aug 16, 2023 at 04:54:54AM -0400, Przemek Kitszel wrote:
> > > > > > > Extend struct ice_vf by vfdev.
> > > > > > > Calculation of vfdev falls more nicely into ice_create_vf_entries().
> > > > > > >
> > > > > > > Caching of vfdev enables simplification of
> > ice_restore_all_vfs_msi_state().
> > > > > >
> > > > > > I see that old code had access to pci_dev * of VF without any locking
> > > > > > from concurrent PCI core access. How is it protected? How do you make
> > > > > > sure that vfdev is valid?
> > > > > >
> > > > > > Generally speaking, it is rarely good idea to cache VF pci_dev pointers
> > > > > > inside driver.
> > > > > >
> > > > > > Thanks
> > > > >
> > > > > Overall, I do agree that ice driver, as a whole, has room for improvement in
> > > > > terms of synchronization, objects lifetime, and similar.
> > > > >
> > > > > In this particular case, I don't see any reason of PCI reconfiguration
> > > > > during VF lifetime, but likely I'm missing something?
> > > >
> > > > You are caching VF pointer in PF,
> > >
> > > that's correct that the driver is PF/ice
> > >
> > > > and you are subjected to PF lifetime
> > > > and not VF lifetime.
> > >
> > > this belongs to struct ice_vf, which should have VF lifetime,
> > > otherwise it's already at risk
> > 
> > I'm not so sure about it. ICE used to use devm_* API and not explicit
> > kalloc/kfree calls, it is not clear anymore the lifetime scope of VF
> > structure.
> > 
> > Thanks
> > 
> 
> The ice_vf structure is now reference counted with a kref, and is created when VFs are added, and removed when the VF is removed.

Ohh, great, thanks

> 
> Thanks,
> Jake
> 
>