> >>> +static int sst26vf_nor_post_sfdp(struct spi_nor *nor)
> >>> +{
> >>> + struct nvmem_device *nvmem;
> >>> +
> >>> + sst26vf_sfdp_nvmem_config.dev = nor->dev;
> >>> + sst26vf_sfdp_nvmem_config.size = nor->sfdp->num_dwords * sizeof(*nor->sfdp->dwords);
> >>> + sst26vf_sfdp_nvmem_config.priv = nor;
> >>> + sst26vf_sfdp_nvmem_config.reg_read = sst26vf_sfdp_mac_addr_read;
> >>> +
> >>> + nvmem = devm_nvmem_register(nor->dev, &sst26vf_sfdp_nvmem_config);
> >>> + if (IS_ERR(nvmem)) {
> >>> + dev_err(nor->dev, "failed to register NVMEM device: %ld\n", PTR_ERR(nvmem));
> >>> + return PTR_ERR(nvmem);
> >>
> >> I don't think it makes sense to have this one-off in a particular
> >> driver. If at all, this should be handled in the core. Sorry, but
> >> this really looks like an ugly hack.
> >>
> >
> > Because the EUI identifier within the SFDP is unique to the
> > SST26VF064BEUI flash, I opted to handle it here rather than in the core.
> >
> > Also here the MAC address data resides within the 0x260-0x26F range, I
> > will resize the nvmem_config.size to 0x10 instead of registering the
> > full SFDP region as NVMEM.
>
> Open question to all parties in this thread: how do we give an offset in
> the device tree that is relative to the sfdp region and not the data
> region? I believe we care not to mix these areas while describing.
You don't do it, because there is not even a relative offset that is
fixed. There should be a pointer to the vendor table inside the SFDP
structure. Thus, you need to properly parse it.
Regarding how to reference it within the device tree, I'd assume
something along 'compatible = "jedec,sfdp-vendor-table-NNN";' or
similar. But no static/relative offsets.
-michael
