On Tue, Jun 24, 2025 at 10:13:38AM -0600, Ahmed Zaki wrote:
> 
> 
> On 2025-06-24 3:40 a.m., Simon Horman wrote:
> > On Mon, Jun 23, 2025 at 09:48:02AM -0600, Ahmed Zaki wrote:
> > > 
> > > 
> > > On 2025-06-21 6:13 a.m., Simon Horman wrote:
> > > > On Fri, Jun 20, 2025 at 11:15:48AM -0600, Ahmed Zaki wrote:
> > > > > The IRQ coalescing config currently reside only inside struct
> > > > > idpf_q_vector. However, all idpf_q_vector structs are de-allocated and
> > > > > re-allocated during resets. This leads to user-set coalesce 
> > > > > configuration
> > > > > to be lost.
> > > > > 
> > > > > Add new fields to struct idpf_vport_user_config_data to save the user
> > > > > settings and re-apply them after reset.
> > > > > 
> > > > > Reviewed-by: Madhu Chittim <[email protected]>
> > > > > Signed-off-by: Ahmed Zaki <[email protected]>
> > > > 
> > > > Hi Ahmed,
> > > > 
> > > > I am wondering if this patch also preserves coalescing settings in the 
> > > > case
> > > > where.
> > > > 
> > > > 1. User sets coalescence for n queues
> > > > 2. The number of queues is reduced, say to m (where m < n)
> > > > 3. The user then increases the number of queues, say back to n
> > > > 
> > > > It seems to me that in this scenario it's reasonable to preserve
> > > > the settings for queues 0 to m, bit not queues m + 1 to n.
> > > 
> > > Hi Simon,
> > > 
> > > I just did a quick test and it seems new settings are preserved in the 
> > > above
> > > scenario: all n queues have the new coalescing settings.
> > 
> > Hi Ahmed,
> > 
> > Thanks for looking into this.
> > 
> > > > But perhaps this point is orthogonal to this change.
> > > > I am unsure.
> > > > 
> > > 
> > > Agreed, but let me know if it is a showstopper.
> > 
> > If preserving the status of all n queues, rather than just the first m
> > queues, in the scenario described above is new behaviour added by this
> > patch then I would lean towards yes. Else no.
> > 
> > 
> 
> I don't believe we can call this new behavior. Actually, the napi IRQ
> affinity pushed to CORE few weeks ago behaves in the same manner; deleting
> queues and re-adding them restores the user-set IRQ affinity.

Right, in that case it's certainly not a showstopper.

Reviewed-by: Simon Horman <[email protected]>

Reply via email to