Re: [PATCH net-next 1/1] ice: Improve MSI-X vector enablement fallback logic

["Venkataramanan, Anirudh" <anirudh.venkataramanan@intel.com>]

On Thu, 2021-01-14 at 16:42 -0800, Jakub Kicinski wrote:
> On Wed, 13 Jan 2021 15:42:26 -0800 Tony Nguyen wrote:
> > From: Brett Creeley <brett.creeley@intel.com>
> > 
> > The current MSI-X enablement logic either tries to enable best-case
> > MSI-X vectors and if that fails we only support a bare-minimum set.
> > This is not very flexible and the current logic is broken when
> > actually
> > allocating and reserving MSI-X in the driver. Fix this by improving
> > the
> > fall-back logic and fixing the allocation/reservation of MSI-X when
> > the
> > best-case MSI-X vectors are not received from the OS.
> > 
> > The new fall-back logic is described below with each [#] being an
> > attempt at enabling a certain number of MSI-X. If any of the steps
> > succeed, then return the number of MSI-X enabled from
> > ice_ena_msix_range(). If any of the attempts fail, then goto the
> > next
> > step.
> > 
> > Attempt [0]: Enable the best-case scenario MSI-X vectors.
> > 
> > Attempt [1]: Enable MSI-X vectors with the number of pf-
> > >num_lan_msix
> > reduced by a factor of 2 from the previous attempt (i.e.
> > num_online_cpus() / 2).
> > 
> > Attempt [2]: Same as attempt [1], except reduce by a factor of 4.
> > 
> > Attempt [3]: Enable the bare-minimum MSI-X vectors.
> > 
> > Also, if the adjusted_base_msix ever hits the minimum required for
> > LAN,
> > then just set the needed MSI-X for that feature to the minimum
> > (similar to attempt [3]).
> 
> I don't really get why you switch to this manual "exponential back-
> off"
> rather than continuing to use pci_enable_msix_range(), but fixing the
> capping to ICE_MIN_LAN_VECS.

As per the current logic, if the driver does not get the number of MSI-
X vectors it needs, it will immediately drop to "Do I have at least two
(ICE_MIN_LAN_VECS) MSI-X vectors?". If yes, the driver will enable a
single Tx/Rx traffic queue pair, bound to one of the two MSI-X vectors.

This is a bit of an all-or-nothing type approach. There's a mid-ground
that can allow more queues to be enabled (ex. driver asked for 300
vectors, but got 68 vectors, so enabled 64 data queues) and this patch
implements the mid-ground logic. 

This mid-ground logic can also be implemented based on the return value
of pci_enable_msix_range() but IMHO the implementation in this patch
using pci_enable_msix_exact is better because it's always only
enabling/reserving as many MSI-X vectors as required, not more, not
less.

> 
> > To fix the allocation/reservation of MSI-X, the PF VSI needs to
> > take
> > into account the pf->num_lan_msix available and only allocate up to
> > that
> > many MSI-X vectors. To do this, limit the number of Tx and Rx
> > queues
> > based on pf->num_lan_msix. This is done because we don't want more
> > than
> > 1 Tx/Rx queue per interrupt due to performance concerns. Also,
> > limit the
> > number of MSI-X based on pf->num_lan_msix available.
> > 
> > Also, prevent users from enabling more combined queues than there
> > are
> > MSI-X available via ethtool.
> 
> Right, this part sounds like a fix, and should be a separate patch
> against net, not net-next.

ACK. We will do a different patch for this.

> 
> > Fixes: 152b978a1f90 ("ice: Rework ice_ena_msix_range")
> > Signed-off-by: Brett Creeley <brett.creeley@intel.com>
> > Tested-by: Tony Brelinski <tonyx.brelinski@intel.com>
> > Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>