Re: [PATCH 6/7] cxl/region: Fix 'distance' calculation with passthrough ports

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Verma, Vishal L wrote:
> On Thu, 2022-11-03 at 17:30 -0700, Dan Williams wrote:
> > When programming port decode targets, the algorithm wants to ensure that
> > two devices are compatible to be programmed as peers beneath a given
> > port. A compatible peer is a target that shares the same dport, and
> > where that target's interleave position also routes it to the same
> > dport. Compatibility is determined by the device's interleave position
> > being >= to distance. For example, if a given dport can only map every
> > Nth position then positions less than N away from the last target
> > programmed are incompatible.
> > 
> > The @distance for the host-bridge-cxl_port a simple dual-ported
>                                    ^
> Is this meant to be "the distance for the host-bridge to a cxl_port"?

No, but I will preface this explanation by admitting "distance" may not
be the best term for what this value is describing. CXL decode routes to
targets in a round robin fashion per-port. Take the diagram below:

 ┌───────────────────────────────────┬──┐
 │WINDOW0                            │x2│
 └─────────┬─────────────────┬───────┴──┘
           │                 │
 ┌─────────▼────┬──┐  ┌──────▼───────┬──┐
 │HB0           │x2│  │HB1           │x2│
 └──┬────────┬──┴──┘  └─┬─────────┬──┴──┘
    │        │          │         │
 ┌──▼─┬──┐ ┌─▼──┬──┐  ┌─▼──┬──┐ ┌─▼──┬──┐
 │DEV0│x4│ │DEV1│x4│  │DEV2│x4│ │DEV3│x4│
 └────┴──┘ └────┴──┘  └────┴──┘ └────┴──┘
    0         2          1         3

...where an x4 region is being established, and all the xN values are
the interleave-ways settings for those ports/devices. The @distance
value for that "HB0" port is 2. I.e. in order for 2 devices in that
region to be mapped under HB0 they must be at position X and X+2 in the
region.  The algorithm needs to be flexible to also allow this
configuration:

 ┌───────────────────────────────────┬──┐
 │WINDOW0                            │x2│
 └─────────┬─────────────────┬───────┴──┘
           │                 │
 ┌─────────▼────┬──┐  ┌──────▼───────┬──┐
 │HB0           │x2│  │HB1           │x2│
 └──┬────────┬──┴──┘  └─┬─────────┬──┴──┘
    │        │          │         │
 ┌──▼─┬──┐ ┌─▼──┬──┐  ┌─▼──┬──┐ ┌─▼──┬──┐
 │DEV3│x4│ │DEV2│x4│  │DEV1│x4│ │DEV0│x4│
 └────┴──┘ └────┴──┘  └────┴──┘ └────┴──┘
    3         1          2         0

...and the algorithm can not know that a device is in the wrong position
until trying to map the peer (like DEV0 and DEV1 are peers) into the
decode. So "The @distance for the host-bridge-cxl_port" is referring to
the value "2" for host-bridge-cxl_port:HB0 and host-bridge-cxl_port:HB1
in the diagram.

> Also missing '/in/ a simple dual-ported..'?

Yes to this fixup though.

> 
> > host-bridge configuration with 2 direct-attached devices is 1. An x2
> > region divded by 2 dports to reach 2 region targets.
> 
> The second sentence seems slightly incomprehensible too.

Oh, I think I meant that to be s/. An/, i.e. an/:

"...host-bridge configuration with 2 direct-attached devices is 1, i.e.
an x2 region divded by 2 dports to reach 2 region positions"


> > 
> > An x4 region under an x2 host-bridge would need 2 intervening switches
> > where the @distance at the host bridge level is 2 (x4 region divided by
> > 2 switches to reach 4 devices).
> > 
> > However, the distance between peers underneath a single ported
> > host-bridge is always zero because there is no limit to the number of
> > devices that can be mapped. In other words, there are no decoders to
> > program in a passthrough, all descendants are mapped and distance only
> > starts matters for the intervening descendant ports of the passthrough
> 
> starts to matter?

s/starts matters/matters/

> 
> > port.
> > 
> > Add tracking for the number of dports mapped to a port, and use that to
> > detect the passthrough case for calculating @distance.
> > 
> > Cc: <stable@xxxxxxxxxxxxxxx>
> > Reported-by: Bobo WL <lmw.bobo@xxxxxxxxx>
> > Reported-by: Jonathan Cameron <Jonathan.Cameron@xxxxxxxxxx>
> > Link: http://lore.kernel.org/r/20221010172057.00001559@xxxxxxxxxx
> > Fixes: 27b3f8d13830 ("cxl/region: Program target lists")
> > Signed-off-by: Dan Williams <dan.j.williams@xxxxxxxxx>
> > ---
> >  drivers/cxl/core/port.c   |   11 +++++++++--
> >  drivers/cxl/core/region.c |    9 ++++++++-
> >  drivers/cxl/cxl.h         |    2 ++
> >  3 files changed, 19 insertions(+), 3 deletions(-)
> 
> Other than the above, looks good,
> 
> Reviewed-by: Vishal Verma <vishal.l.verma@xxxxxxxxx>
> 
> > 
> > diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c
> > index bffde862de0b..e7556864ea80 100644
> > --- a/drivers/cxl/core/port.c
> > +++ b/drivers/cxl/core/port.c
> > @@ -811,6 +811,7 @@ static struct cxl_dport *find_dport(struct cxl_port *port, int id)
> >  static int add_dport(struct cxl_port *port, struct cxl_dport *new)
> >  {
> >         struct cxl_dport *dup;
> > +       int rc;
> >  
> >         device_lock_assert(&port->dev);
> >         dup = find_dport(port, new->port_id);
> > @@ -821,8 +822,14 @@ static int add_dport(struct cxl_port *port, struct cxl_dport *new)
> >                         dev_name(dup->dport));
> >                 return -EBUSY;
> >         }
> > -       return xa_insert(&port->dports, (unsigned long)new->dport, new,
> > -                        GFP_KERNEL);
> > +
> > +       rc = xa_insert(&port->dports, (unsigned long)new->dport, new,
> > +                      GFP_KERNEL);
> > +       if (rc)
> > +               return rc;
> > +
> > +       port->nr_dports++;
> > +       return 0;
> >  }
> >  
> >  /*
> > diff --git a/drivers/cxl/core/region.c b/drivers/cxl/core/region.c
> > index c52465e09f26..c0253de74945 100644
> > --- a/drivers/cxl/core/region.c
> > +++ b/drivers/cxl/core/region.c
> > @@ -990,7 +990,14 @@ static int cxl_port_setup_targets(struct cxl_port *port,
> >         if (cxl_rr->nr_targets_set) {
> >                 int i, distance;
> >  
> > -               distance = p->nr_targets / cxl_rr->nr_targets;
> > +               /*
> > +                * Passthrough ports impose no distance requirements between
> > +                * peers
> > +                */
> > +               if (port->nr_dports == 1)
> > +                       distance = 0;
> > +               else
> > +                       distance = p->nr_targets / cxl_rr->nr_targets;
> >                 for (i = 0; i < cxl_rr->nr_targets_set; i++)
> >                         if (ep->dport == cxlsd->target[i]) {
> >                                 rc = check_last_peer(cxled, ep, cxl_rr,
> > diff --git a/drivers/cxl/cxl.h b/drivers/cxl/cxl.h
> > index 1164ad49f3d3..ac75554b5d76 100644
> > --- a/drivers/cxl/cxl.h
> > +++ b/drivers/cxl/cxl.h
> > @@ -457,6 +457,7 @@ struct cxl_pmem_region {
> >   * @regions: cxl_region_ref instances, regions mapped by this port
> >   * @parent_dport: dport that points to this port in the parent
> >   * @decoder_ida: allocator for decoder ids
> > + * @nr_dports: number of entries in @dports
> >   * @hdm_end: track last allocated HDM decoder instance for allocation ordering
> >   * @commit_end: cursor to track highest committed decoder for commit ordering
> >   * @component_reg_phys: component register capability base address (optional)
> > @@ -475,6 +476,7 @@ struct cxl_port {
> >         struct xarray regions;
> >         struct cxl_dport *parent_dport;
> >         struct ida decoder_ida;
> > +       int nr_dports;
> >         int hdm_end;
> >         int commit_end;
> >         resource_size_t component_reg_phys;
> > 
> 






[Index of Archives]     [Linux Kernel]     [Kernel Development Newbies]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite Hiking]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux