Re: [PATCH v3 4/9] cxl/mem: Skip intermediate port enumeration of restricted endpoints (RCDs)

[Robert Richter <rrichter@amd.com>]

On 15.11.22 10:08:51, Dan Williams wrote:
> Robert Richter wrote:
> > On 14.11.22 16:07:58, Dan Williams wrote:
> > > Robert Richter wrote:
> > > > When an endpoint is found, all ports in beetween the endpoint and the
> > > > CXL host bridge need to be created. In the RCH case there are no ports
> > > > in between a host bridge and the endpoint. Skip the enumeration of
> > > > intermediate ports.
> > > > 
> > > > The port enumeration does not only create all ports, it also
> > > > initializes the endpoint chain by adding the endpoint to every
> > > > downstream port up to the root bridge. This must be done also in RCD
> > > > mode, but is much more simple as the endpoint only needs to be added
> > > > to the host bridge's dport.
> > > > 
> > > > Note: For endpoint removal the cxl_detach_ep() is not needed as it is
> > > > released in cxl_port_release().
> > > > 
> > > > Signed-off-by: Robert Richter <rrichter@amd.com>
> > > > ---
> > > >  drivers/cxl/core/port.c | 18 ++++++++++++++++--
> > > >  1 file changed, 16 insertions(+), 2 deletions(-)
> > > > 
> > > > diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c
> > > > index d10c3580719b..e21a9c3fe4da 100644
> > > > --- a/drivers/cxl/core/port.c
> > > > +++ b/drivers/cxl/core/port.c
> > > > @@ -1366,8 +1366,24 @@ int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)
> > > >  {
> > > >  	struct device *dev = &cxlmd->dev;
> > > >  	struct device *iter;
> > > > +	struct cxl_dport *dport;
> > > > +	struct cxl_port *port;
> > > >  	int rc;
> > > >  
> > > > +	/*
> > > > +	 * Skip intermediate port enumeration in the RCH case, there
> > > > +	 * are no ports in between a host bridge and an endpoint. Only
> > > > +	 * initialize the EP chain.
> > > > +	 */
> > > > +	if (is_cxl_restricted(cxlmd)) {
> > > 
> > > I changed this to:
> > > 
> > > 	if (cxlmd->cxlds->rcd) {
> > 
> > I a mail to Bjorn I suggested to have cxl_dev_cap and a cxl_port_cap
> > in the pci_dev struct that could be looked up too including RCD mode.
> > Checking the pci_dev looks more reasonable to me here, though we could
> > have a flag of it in cxlds too.
> 
> Would that not need the PCI core to understand how to walk the RCRB
> generically? As far as I understand the RCRB association is ACPI.CEDT
> specific.

I am thinking of doing some sort of cxl_setup_pci_dev(pdev) in
cxl_pci_probe() which extracts the caps and writes them into a struct
pci_dev. Possibly the CXL 68B Flit and VH Capable/Enable bit could be
used for RCD mode. But if that is not feasible for some reason a flag
somewhere could work too.

-Robert