Re: [PATCH v16 06/22] sfc: make regs setup with checking and set media ready

[Dan Williams <dan.j.williams@intel.com>]

Alejandro Lucero Palau wrote:
> 
> On 5/21/25 19:34, Dan Williams wrote:
> > alejandro.lucero-palau@ wrote:
> >> From: Alejandro Lucero <alucerop@amd.com>
> >>
> >> Use cxl code for registers discovery and mapping.
> >>
> >> Validate capabilities found based on those registers against expected
> >> capabilities.
> >>
> >> Set media ready explicitly as there is no means for doing so without
> >> a mailbox and without the related cxl register, not mandatory for type2.
> >>
> >> Signed-off-by: Alejandro Lucero <alucerop@amd.com>
> >> Reviewed-by: Martin Habets <habetsm.xilinx@gmail.com>
> >> Reviewed-by: Zhi Wang <zhi@nvidia.com>
> >> Acked-by: Edward Cree <ecree.xilinx@gmail.com>
> >> Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
> >> Reviewed-by: Ben Cheatham <benjamin.cheatham@amd.com>
> >> ---
> >>   drivers/net/ethernet/sfc/efx_cxl.c | 26 ++++++++++++++++++++++++++
> >>   1 file changed, 26 insertions(+)
> >>
> >> diff --git a/drivers/net/ethernet/sfc/efx_cxl.c b/drivers/net/ethernet/sfc/efx_cxl.c
> >> index 753d5b7d49b6..e94af8bf3a79 100644
> >> --- a/drivers/net/ethernet/sfc/efx_cxl.c
> >> +++ b/drivers/net/ethernet/sfc/efx_cxl.c
> >> @@ -19,10 +19,13 @@
> >>   
> >>   int efx_cxl_init(struct efx_probe_data *probe_data)
> >>   {
> >> +	DECLARE_BITMAP(expected, CXL_MAX_CAPS) = {};
> >> +	DECLARE_BITMAP(found, CXL_MAX_CAPS) = {};
> >>   	struct efx_nic *efx = &probe_data->efx;
> >>   	struct pci_dev *pci_dev = efx->pci_dev;
> >>   	struct efx_cxl *cxl;
> >>   	u16 dvsec;
> >> +	int rc;
> >>   
> >>   	probe_data->cxl_pio_initialised = false;
> >>   
> >> @@ -43,6 +46,29 @@ int efx_cxl_init(struct efx_probe_data *probe_data)
> >>   	if (!cxl)
> >>   		return -ENOMEM;
> >>   
> >> +	set_bit(CXL_DEV_CAP_HDM, expected);
> >> +	set_bit(CXL_DEV_CAP_RAS, expected);
> >> +
> >> +	rc = cxl_pci_accel_setup_regs(pci_dev, &cxl->cxlds, found);
> >> +	if (rc) {
> >> +		pci_err(pci_dev, "CXL accel setup regs failed");
> >> +		return rc;
> >> +	}
> >> +
> >> +	/*
> >> +	 * Checking mandatory caps are there as, at least, a subset of those
> >> +	 * found.
> >> +	 */
> >> +	if (cxl_check_caps(pci_dev, expected, found))
> >> +		return -ENXIO;
> > This all looks like an obfuscated way of writing:
> >
> >      cxl_pci_setup_regs(pdev, CXL_REGLOC_RBI_MEMDEV, &map);
> >      if (!map.component_map.ras.valid || !map.component_map.hdm_decoder.valid)
> >           /* sfc cxl expectations not met */
> 
> 
> That is an unfair comment.
> 
> 
> Map is not available here 

Why is @map not available? It was made public in patch1?

> and I do not think it should.

...that is the point of contention.

> The CXL API 
> should hide all this.

A new "capabilities" contract is not hiding anything, it is layering on
a redundant mechanism.

> Adding that new accel function avoids repeating something all the
> drivers will go through:

When / if multiple drivers end up with the same policy, then look to
refactor them into a helper for that policy.

> cxl_pci_setup_regs(pdev, CXL_REGLOC_RBI_MEMDEV, &map);
> 
> Maybe cxl_map_device_regs(&map, &cxlds->regs.device_regs);
> 
> cxl_pci_setup_regs(pdev, CXL_REGLOC_RBI_COMPONENT, &cxlds->reg_map, caps);
> 
> And maybe cxl_map_component_regs(&cxlds->reg_map, &cxlds->regs.component, BIT(CXL_CM_CAP_CAP_ID_RAS));

This still looks like no net improvement over the mechanisms the core
currently has. cxl_pci_accel_setup_memdev_regs() spends time explaining
how it might not work for all CXL accelerator device types, it looks
like pure maintenance burden to me. I want future accelerator driver
writers to carefully think through this problem and not look to a
cxl_pci_accel_ helper to do that for them. Outside of Type-3 there is
simply too much freedom in the CXL specification for a client driver to
expect a pre-built helper for their use case.