Linux 4.5's nfp_net_irq_unmask_msix() vs Xen

Linux 4.5's nfp_net_irq_unmask_msix() vs Xen

[Jan Beulich]

All,

the new driver homed under drivers/net/ethernet/netronome/nfp/
has some direct MSI-X table manipulation which quite clearly is
incompatible with Xen. According to the comment preceding the
function, bypassing the Linux IRQ subsystem is intentional here.
Irrespective of the question of whether that's a good idea, this
also bypassing the abstractions allowing MSI to work on Xen
clearly needs addressing. Does anyone have any thoughts on how
to reasonably achieve this? Calling pci_msi_unmask_irq() would
seem to be an option (the symbol at least is exported), but likely
isn't intended to be used that way.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: Linux 4.5's nfp_net_irq_unmask_msix() vs Xen

[Jakub Kicinski]

Hello Jan.

On Tue, 22 Mar 2016 05:29:54 -0600, Jan Beulich wrote:
> All,
> 
> the new driver homed under drivers/net/ethernet/netronome/nfp/
> has some direct MSI-X table manipulation which quite clearly is
> incompatible with Xen. According to the comment preceding the
> function, bypassing the Linux IRQ subsystem is intentional here.
> Irrespective of the question of whether that's a good idea, this
> also bypassing the abstractions allowing MSI to work on Xen
> clearly needs addressing. Does anyone have any thoughts on how
> to reasonably achieve this? Calling pci_msi_unmask_irq() would
> seem to be an option (the symbol at least is exported), but likely
> isn't intended to be used that way.

Did you check the driver code before sending this?  I think you may
be referring to something which was removed months ago (and never
upstreamed).

Kuba

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: Linux 4.5's nfp_net_irq_unmask_msix() vs Xen

[Jakub Kicinski]

On Tue, 22 Mar 2016 11:34:25 +0000, Jakub Kicinski wrote:
> Hello Jan.
> 
> On Tue, 22 Mar 2016 05:29:54 -0600, Jan Beulich wrote:
> > All,
> > 
> > the new driver homed under drivers/net/ethernet/netronome/nfp/
> > has some direct MSI-X table manipulation which quite clearly is
> > incompatible with Xen. According to the comment preceding the
> > function, bypassing the Linux IRQ subsystem is intentional here.
> > Irrespective of the question of whether that's a good idea, this
> > also bypassing the abstractions allowing MSI to work on Xen
> > clearly needs addressing. Does anyone have any thoughts on how
> > to reasonably achieve this? Calling pci_msi_unmask_irq() would
> > seem to be an option (the symbol at least is exported), but likely
> > isn't intended to be used that way.
> 
> Did you check the driver code before sending this?  I think you may
> be referring to something which was removed months ago (and never
> upstreamed).

Oh, sorry.  I think you just mean unmasking automasked interrupts.  Is
that correct?  We used to do more MSI-X magic...

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: Linux 4.5's nfp_net_irq_unmask_msix() vs Xen

[Jan Beulich]

>>> On 22.03.16 at 12:38, <jakub.kicinski@netronome.com> wrote:
> On Tue, 22 Mar 2016 11:34:25 +0000, Jakub Kicinski wrote:
>> On Tue, 22 Mar 2016 05:29:54 -0600, Jan Beulich wrote:
>> > the new driver homed under drivers/net/ethernet/netronome/nfp/
>> > has some direct MSI-X table manipulation which quite clearly is
>> > incompatible with Xen. According to the comment preceding the
>> > function, bypassing the Linux IRQ subsystem is intentional here.
>> > Irrespective of the question of whether that's a good idea, this
>> > also bypassing the abstractions allowing MSI to work on Xen
>> > clearly needs addressing. Does anyone have any thoughts on how
>> > to reasonably achieve this? Calling pci_msi_unmask_irq() would
>> > seem to be an option (the symbol at least is exported), but likely
>> > isn't intended to be used that way.
>> 
>> Did you check the driver code before sending this?  I think you may
>> be referring to something which was removed months ago (and never
>> upstreamed).
> 
> Oh, sorry.  I think you just mean unmasking automasked interrupts.  Is
> that correct?

Yes (see $subject).

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel