Re: [PATCH v4 12/12] NTB: epf: Implement db_vector_count/mask for doorbells

On Tue, Jun 23, 2026 at 11:31:34AM -0500, Bjorn Helgaas wrote:
> On Wed, May 13, 2026 at 11:49:23AM +0900, Koichiro Den wrote:
> > Implement .db_vector_count and .db_vector_mask so ntb core/clients can
> > map doorbell events to per-vector work.
> > 
> > Report vectors as 0..(db_count - 2) (skipping the unused slot) and return
> > BIT_ULL(db_vector) for the corresponding doorbell bit. Use
> > ntb_epf_db_vector_count() for bounds checks in ntb_epf_db_vector_mask(), so
> > the same lower-bound guard is applied before building the bitmask.
> > 
> > Signed-off-by: Koichiro Den <[email protected]>
> > ---
> > Changes since v3:
> >   - Reuse ntb_epf_db_vector_count() from ntb_epf_db_vector_mask() for bounds.
> >   - Return 0 when db_count is below NTB_EPF_MIN_DB_COUNT.
> >   - Drop Reviewed-by tags due to the changes.
> > 
> >  drivers/ntb/hw/epf/ntb_hw_epf.c | 32 ++++++++++++++++++++++++++++++++
> >  1 file changed, 32 insertions(+)
> > 
> > diff --git a/drivers/ntb/hw/epf/ntb_hw_epf.c b/drivers/ntb/hw/epf/ntb_hw_epf.c
> > index 10618e462229..af5755472842 100644
> > --- a/drivers/ntb/hw/epf/ntb_hw_epf.c
> > +++ b/drivers/ntb/hw/epf/ntb_hw_epf.c
> > @@ -434,6 +434,36 @@ static u64 ntb_epf_db_valid_mask(struct ntb_dev *ntb)
> >  	return ntb_ndev(ntb)->db_valid_mask;
> >  }
> >  
> > +static int ntb_epf_db_vector_count(struct ntb_dev *ntb)
> > +{
> > +	struct ntb_epf_dev *ndev = ntb_ndev(ntb);
> > +	unsigned int db_count = ndev->db_count;
> > +
> > +	/*
> > +	 * db_count includes an extra skipped slot due to the legacy
> > +	 * doorbell layout. Expose only the real doorbell vectors.
> > +	 */
> > +	if (db_count < NTB_EPF_MIN_DB_COUNT)
> > +		return 0;
> > +
> > +	return db_count - 1;
> > +}
> > +
> > +static u64 ntb_epf_db_vector_mask(struct ntb_dev *ntb, int db_vector)
> > +{
> > +	int nr_vec;
> > +
> > +	/*
> > +	 * db_count includes one skipped slot in the legacy layout. Valid
> > +	 * doorbell vectors are therefore [0 .. (db_count - 2)].
> > +	 */
> > +	nr_vec = ntb_epf_db_vector_count(ntb);
> > +	if (db_vector < 0 || db_vector >= nr_vec)
> > +		return 0;
> > +
> > +	return BIT_ULL(db_vector);
> 

Hi Bjorn,

> This all ends up looking slightly different from pci-epf-vntb.c, and
> it's not obvious whether the differences are essential or superfluous.

I don't think there is an actual bug here.

The thing is that "db_count" has meant slightly different things in
pci-epf-vntb.c and ntb_hw_epf.c since their introduction.

In pci-epf-vntb.c, db_count is the total number including the link slot and the
historically unused/skipped slot. It is also exposed via configfs, so changing
that meaning now would affect existing users' configfs settings.

So vntb_epf_db_vector_count() subtracts EPF_IRQ_DB_START because the NTB core
should only see the usable doorbell vectors via .db_vector_count().

> 
> pci-epf-vntb.c:
> 
>   enum EPF_IRQ_DB_START                           # value 2
>   #define MIN_DB_COUNT  (EPF_IRQ_DB_START + 1)    # value 3
> 
>   vntb_epf_db_vector_count() comment says "db_count is total number of
>   doorbell slots exposed to peer, including reserved ones".  But it
>   returns "db_count - EPF_IRQ_DB_START", which suggests that the
>   reserves slots are not exposed?
> 
>   vntb_epf_db_vector_count() comment lists two reserved slots
> 
>   vntb_epf_db_vector_count() returns "db_count - EPF_IRQ_DB_START"
> 
>   vntb_epf_db_vector_mask() comment mentions "two reserved slots"
> 
> ntb_hw_epf.c:
> 
>   enum EPF_IRQ_DB_START                # value 2
>   #define NTB_EPF_MIN_DB_COUNT  3      # value 3, not based on EPF_IRQ_DB_START
> 
>   ntb_epf_db_vector_count() comment says "expose only real doorbell
>   vectors".  I guess this *excludes* the reserved ones?

Yes.

> 
>   ntb_epf_db_vector_count() comment mentions "an extra skipped slot"
>   but apparently actually accounts for *two* slots
>   (NTB_EPF_MIN_DB_COUNT==3)
> 
>   ntb_epf_db_vector_count() returns "db_count - 1"   # should it be -2?

This is intentionally -1.

ntb_hw_epf sets "ndev->db_count = irq - 1" where the "-1" accounts for the link
slot. That goes back to the initial ntb_hw_epf support commit 812ce2f8d14ea. So
the number of usable doorbell vectors is "db_count - 1", subtracting the
remaining unused/skipped slot.

> 
>   ntb_epf_db_vector_mask() comment mentions "one skipped slot"
> 
> I don't want to change anything during the v7.2 merge window, but if
> there are any actual bugs here, they could be fixed after v7.2-rc1.
> If there are no bugs but this could be clarified to reduce confusion,
> we could do that for v7.3.

If preferred, I can send a follow-up patch that adds comments to make the
distinction clearer.

Best regards,
Koichiro