[Skiboot] [PATCH v2 5/5] hdata: Parse SPD data

[Stewart Smith]

Vasant Hegde <hegdevasant at linux.vnet.ibm.com> writes:
> Parse SPD data and populate device tree.
>
> list of properites parsing from SPD:
> -----------------------------------
> [root at ltc-wspoon dimm at d00f]# lsprop .
> memory-id        0000000c (12)   <-- DIMM type
> device_type      "memory-dimm-ddr4"
> serial           15d9ad22 (366587170)
> part-number      "36ASF2G72PZ-2G6B2   "
> manufacturer-id  0000802c (32812)   <-- Vendor ID, we can get vendor name from this ID
>
> Signed-off-by: Vasant Hegde <hegdevasant at linux.vnet.ibm.com>
> ---
>  hdata/memory.c | 43 ++++++++++++++++++++++++++++++++++++++++++-
>  1 file changed, 42 insertions(+), 1 deletion(-)

Does this match well enough whats in doc/device-tree/vpd.rst ? There
should probably be a couple of updates there, as it'll mean it's no
longer FSP machine specific :)

>
> diff --git a/hdata/memory.c b/hdata/memory.c
> index d1570eb..956e0d7 100644
> --- a/hdata/memory.c
> +++ b/hdata/memory.c
> @@ -319,16 +319,49 @@ static void vpd_add_ram_area(const struct HDIF_common_hdr *msarea)
>  	}
>  }
>
> +static void vpd_parse_spd(struct dt_node *dimm, const char *spd, u32 size)
> +{
> +	u16 *vendor;
> +	u32 *sn;
> +
> +	/* SPD is too small */
> +	if (size < 512) {
> +		prlog(PR_WARNING, "MSVPD: Invalid SPD size. "
> +		      "Expected 512 byts, got %d\n", size);
> +		return;
> +	}
> +
> +	/* Supports DDR4 format pasing only */
> +	if (spd[0x2] < 0xc) {
> +		prlog(PR_WARNING,
> +		      "MSVPD: SPD format (%x) not supported\n", spd[0x2]);
> +		return;
> +	}

For both of these we should probably have FWTS annotations explaining
what's wrong.

> +
> +	dt_add_property_string(dimm, "device_type", "memory-dimm-ddr4");
> +
> +	dt_add_property_cells(dimm, "memory-id", be16_to_cpu(spd[0x2]));
> +
> +	sn = (u32 *)&spd[0x145];
> +	dt_add_property_cells(dimm, "serial", be32_to_cpu(*sn));
> +
> +	dt_add_property_nstr(dimm, "part-number", &spd[0x149], 20);
> +
> +	vendor = (u16 *)&spd[0x140];
> +	dt_add_property_cells(dimm, "manufacturer-id", be16_to_cpu(*vendor));
> +}

Do we have a good way to link back DIMM to memory address?

-- 
Stewart Smith
OPAL Architect, IBM.