Hi Marek,
> Subject: Re: [PATCH v5 1/2] arm: rmobile: Add RZ/G2[HMNE] SoC support
>
> On 10/8/20 10:59 AM, Biju Das wrote:
>
> [...]
>
> > +static const struct udevice_id tfa_cpu_info[] = {
> > + { .compatible = "renesas,r8a774a1", .data =
> RMOBILE_CPU_TYPE_R8A774A1 },
> > + { .compatible = "renesas,r8a774b1", .data =
> RMOBILE_CPU_TYPE_R8A774B1 },
> > + { .compatible = "renesas,r8a774c0", .data =
> RMOBILE_CPU_TYPE_R8A774C0 },
> > + { .compatible = "renesas,r8a774e1", .data =
> RMOBILE_CPU_TYPE_R8A774E1 },
> > + { },
> > +};
> > +
> > +static const u32 get_cpu_type(u32 soc_id) {
> > + const struct udevice_id *of_match = tfa_cpu_info;
> > + int i;
>
> Isn't there already a function in U-Boot which can match on a table of OF
> compatible strings ? I suspect when drivers match on their table of OF
> compatible strings, somesuch function must be used.
of_match function for matching device compatible string. But it doesn't match
SoC compatible string.
So I added a helper function is of_match_node[1] to search from rootnode to
find the soc compatible string.
[1]
http://patchwork.ozlabs.org/project/uboot/patch/20201030140303.11773-1-biju.das...@bp.renesas.com/
Apart from this, This helper function can be used to replace the below codes in
u-boot tree [2] [3]
[2]
https://elixir.bootlin.com/u-boot/latest/source/drivers/serial/serial_uniphier.c#L129
[3]
https://elixir.bootlin.com/u-boot/latest/source/drivers/usb/phy/rockchip_usb2_phy.c#L77
> > + for (i = 0; i < ARRAY_SIZE(tfa_cpu_info); i++) {
> > + if (soc_id == (of_match->data & PRODUCT_MASK) &&
> > + of_machine_is_compatible(of_match->compatible))
> > + return of_match->data;
> > + of_match++;
> > + }
> > +
> > + return soc_id;
> > +}
> > +
> > static u32 rmobile_get_prr(void)
> > {
> > #ifdef CONFIG_RCAR_GEN3
> > @@ -23,7 +48,9 @@ static u32 rmobile_get_prr(void)
> >
> > u32 rmobile_get_cpu_type(void)
> > {
> > - return (rmobile_get_prr() & 0x00007F00) >> 8;
> > + const u32 soc_id = (rmobile_get_prr() & 0x00007F00) >> 8;
>
> The soc_id = ... can be inlined into get_cpu_type().
>
> However, you might want to cache the result of rmobile_get_cpu_type() ,
> because doing OF match every time this is called is expensive.
I agree calling OF match is expensive. So I have ported Renesas SoC
identification driver from Linux to u-boot
which will cache the family type, soc_id and revision. I already sent a patch
for supporting soc_id in UCLASS_SOC in ML[4]
[4]
http://patchwork.ozlabs.org/project/uboot/patch/20201030140724.12773-1-biju.das...@bp.renesas.com/
On the next version, I will send Renesas SoC identification driver, which
supports caching family type which
can be used to provide unique identification for CPU type.
> > +
> > + return get_cpu_type(soc_id);
> > }
>
> [...]
>
> > +/* CPU IDs */
> > +#define RMOBILE_CPU_TYPE_SH73A0 (SOC_ID_SH73A0)
> > +#define RMOBILE_CPU_TYPE_R8A7740 (SOC_ID_R8A7740)
> > +#define RMOBILE_CPU_TYPE_R8A774A1 (SOC_ID_R8A774A1 |
> RZG_CPU_MASK)
> > +#define RMOBILE_CPU_TYPE_R8A774B1 (SOC_ID_R8A774B1 |
> RZG_CPU_MASK)
> > +#define RMOBILE_CPU_TYPE_R8A774C0 (SOC_ID_R8A774C0 |
> RZG_CPU_MASK)
> > +#define RMOBILE_CPU_TYPE_R8A774E1 (SOC_ID_R8A774E1 |
> RZG_CPU_MASK)
> > +#define RMOBILE_CPU_TYPE_R8A7790 (SOC_ID_R8A7790)
> > +#define RMOBILE_CPU_TYPE_R8A7791 (SOC_ID_R8A7791)
> > +#define RMOBILE_CPU_TYPE_R8A7792 (SOC_ID_R8A7792)
> > +#define RMOBILE_CPU_TYPE_R8A7793 (SOC_ID_R8A7793)
> > +#define RMOBILE_CPU_TYPE_R8A7794 (SOC_ID_R8A7794)
> > +#define RMOBILE_CPU_TYPE_R8A7795 (SOC_ID_R8A7795)
> > +#define RMOBILE_CPU_TYPE_R8A7796 (SOC_ID_R8A7796)
> > +#define RMOBILE_CPU_TYPE_R8A77965 (SOC_ID_R8A77965)
> > +#define RMOBILE_CPU_TYPE_R8A77970 (SOC_ID_R8A77970)
> > +#define RMOBILE_CPU_TYPE_R8A77980 (SOC_ID_R8A77980)
> > +#define RMOBILE_CPU_TYPE_R8A77990 (SOC_ID_R8A77990)
> > +#define RMOBILE_CPU_TYPE_R8A77995 (SOC_ID_R8A77995)
>
> The () parentheses are not needed. Also, is there any need for this renaming
> of SOC_ID_R8... to RMOBILE_CPU... ?
OK, will take out parenthesis. RMOBILE_CPU_TYPE is unique, where SOC_ID_XXXX is
not unique.
> > +/* RZ/G CPU Identification Mask */
> > +#define RZG_CPU_MASK 0x1000
>
> This mask should have a comment which explicitly states that it is not related
> to the PRR register content.
Ok. Agreed.
Thanks and regards,
Biju
