Date:   Wed, 22 Mar 2023 19:40:41 +0100
From:   Piotr Raczynski <piotr.raczynski@...el.com>
To:     Jacob Keller <jacob.e.keller@...el.com>
CC:     <intel-wired-lan@...ts.osuosl.org>, <netdev@...r.kernel.org>,
        <michal.swiatkowski@...el.com>, <shiraz.saleem@...el.com>,
        <sridhar.samudrala@...el.com>, <jesse.brandeburg@...el.com>,
        <aleksander.lobakin@...el.com>, <lukasz.czapnik@...el.com>
Subject: Re: [PATCH net-next v2 0/8] ice: support dynamic interrupt allocation

On Wed, Mar 22, 2023 at 10:33:31AM -0700, Jacob Keller wrote:
> 
> 
> On 3/22/2023 9:25 AM, Piotr Raczynski wrote:
> > This patchset reimplements MSIX interrupt allocation logic to allow dynamic
> > interrupt allocation after MSIX has been initially enabled. This allows
> > current and future features to allocate and free interrupts as needed and
> > will help to drastically decrease number of initially preallocated
> > interrupts (even down to the API hard limit of 1). Although this patchset
> > does not change behavior in terms of actual number of allocated interrupts
> > during probe, it will be subject to change.
> > 
> > First few patches prepares to introduce dynamic allocation by moving
> > interrupt allocation code to separate file and update allocation API used
> > in the driver to the currently preferred one.
> > 
> > Due to the current contract between ice and irdma driver which is directly
> > accessing msix entries allocated by ice driver, even after moving away from
> > older pci_enable_msix_range function, still keep msix_entries array for
> > irdma use.
> > 
> > Next patches refactors and removes redundant code from SRIOV related logic
> > as it also make it easier to move away from static allocation scheme.
> > 
> > Last patches actually enables dynamic allocation of MSIX interrupts. First,
> > introduce functions to allocate and free interrupts individually. This sets
> > ground for the rest of the changes even if that patch still allocates the
> > interrupts from the preallocated pool. Since this patch starts to keep
> > interrupt details in ice_q_vector structure we can get rid of functions
> > that calculates base vector number and register offset for the interrupt
> > as it is equal to the interrupt index. Only keep separate register offset
> > functions for the VF VSIs.
> > 
> > Next, replace homegrown interrupt tracker with much simpler xarray based
> > approach. As new API always allocate interrupts one by one, also track
> > interrupts in the same manner.
> > 
> > Lastly, extend the interrupt tracker to deal both with preallocated and
> > dynamically allocated vectors and use pci_msix_alloc_irq_at and
> > pci_msix_free_irq functions. Since not all architecture supports dynamic
> > allocation, check it before trying to allocate a new interrupt.
> > 
> > As previously mentioned, this patchset does not change number of initially
> > allocated interrupts during init phase but now it can and will likely be
> > changed.
> > 
> > Patch 1-3 -> move code around and use newer API
> > Patch 4-5 -> refactor and remove redundant SRIOV code
> > Patch 6   -> allocate every interrupt individually
> > Patch 7   -> replace homegrown interrupt tracker with xarray
> > Patch 8   -> allow dynamic interrupt allocation
> > 
> > Change history:
> > v1 -> v2:
> > - ice: refactor VF control VSI interrupt handling
> >   - move ice_get_vf_ctrl_vsi to ice_lib.c (ice_vf_lib.c depends on
> >     CONFIG_PCI_IOV)
> > 
> 
> The other option would have been to make ice_vf_lib.h have a no-op
> function that always returned NULL, since we generally would know that
> there are no VF ctrl VSI if CONFIG_PCI_IOV is disabled.
> 
> But I'm ok with it being in ice_lib.c too.
> 
> Thanks,
> Jake

Thanks, that makes more sense, a little bit too hasty here.

Powered by blists - more mailing lists