On Mon, Nov 27, 2023 at 7:31 PM Jonas Karlman jonas@kwiboo.se wrote:
Hi Quentin,
On 2023-11-27 19:42, Quentin Schulz wrote:
Hi Jonas,
From: U-Boot u-boot-bounces@lists.denx.de on behalf of Jonas Karlman jonas@kwiboo.se Sent: 26 November 2023 20:46 To: Kever Yang; Simon Glass; Philipp Tomsich Cc: Jonas Karlman; u-boot@lists.denx.de Subject: [PATCH] rockchip: rk3328: Set efuse auto mode and timing control
Reading from efuse return zero when mainline TF-A is used.
=> dump_efuse 00000000: 00 00 00 00 .... 00000004: 00 00 00 00 .... 00000008: 00 00 00 00 .... 0000000c: 00 00 00 00 .... 00000010: 00 00 00 00 .... 00000014: 00 00 00 00 .... 00000018: 00 00 00 00 .... 0000001c: 00 00 00 00 ....
However, when vendor TF-A blobs is used reading from efuse works.
Change to use auto mode, enable finish and auto access err interrupts and set timing control using same values that vendor TF-A blob use to fix this.
With this efuse can be read when either of mainline TF-A or vendor blob is used.
=> dump_efuse 00000000: 52 4b 33 82 RK3. 00000004: 00 fe 21 55 ..!U 00000008: 52 4b 57 34 RKW4 0000000c: 35 30 32 39 5029 00000010: 00 00 00 00 .... 00000014: 08 25 0c 0f .%.. 00000018: 02 0d 08 00 .... 0000001c: 00 00 f0 00 ....
Signed-off-by: Jonas Karlman jonas@kwiboo.se
arch/arm/mach-rockchip/rk3328/rk3328.c | 34 ++++++++++++++++++++++++++ 1 file changed, 34 insertions(+)
diff --git a/arch/arm/mach-rockchip/rk3328/rk3328.c b/arch/arm/mach-rockchip/rk3328/rk3328.c index de17b8868273..edb22a58fc67 100644 --- a/arch/arm/mach-rockchip/rk3328/rk3328.c +++ b/arch/arm/mach-rockchip/rk3328/rk3328.c @@ -19,6 +19,22 @@ DECLARE_GLOBAL_DATA_PTR; #define GRF_BASE 0xFF100000 #define UART2_BASE 0xFF130000 #define FW_DDR_CON_REG 0xFF7C0040 +#define EFUSE_NS_BASE 0xFF260000
+#define EFUSE_MOD 0x0000 +#define EFUSE_INT_CON 0x0014 +#define EFUSE_T_CSB_P 0x0028 +#define EFUSE_T_PGENB_P 0x002C +#define EFUSE_T_LOAD_P 0x0030 +#define EFUSE_T_ADDR_P 0x0034 +#define EFUSE_T_STROBE_P 0x0038 +#define EFUSE_T_CSB_R 0x003C +#define EFUSE_T_PGENB_R 0x0040 +#define EFUSE_T_LOAD_R 0x0044 +#define EFUSE_T_ADDR_R 0x0048 +#define EFUSE_T_STROBE_R 0x004C
+#define EFUSE_TIMING(s, l) (((s) << 16) | (l))
const char * const boot_devices[BROM_LAST_BOOTSOURCE + 1] = { [BROM_BOOTSOURCE_EMMC] = "/mmc@ff520000", @@ -54,6 +70,24 @@ int arch_cpu_init(void)
/* Disable the ddr secure region setting to make it non-secure */ rk_setreg(FW_DDR_CON_REG, 0x200);
/* Use efuse auto mode */writel(0x46, EFUSE_NS_BASE + EFUSE_MOD);/* Enable efuse finish and auto access err interrupt */writel(0x07, EFUSE_NS_BASE + EFUSE_INT_CON);/* Set efuse timing control */writel(EFUSE_TIMING(1, 241), EFUSE_NS_BASE + EFUSE_T_CSB_P);writel(EFUSE_TIMING(1, 241), EFUSE_NS_BASE + EFUSE_T_PGENB_P);writel(EFUSE_TIMING(1, 241), EFUSE_NS_BASE + EFUSE_T_LOAD_P);writel(EFUSE_TIMING(1, 241), EFUSE_NS_BASE + EFUSE_T_ADDR_P);writel(EFUSE_TIMING(2, 240), EFUSE_NS_BASE + EFUSE_T_STROBE_P);writel(EFUSE_TIMING(1, 4), EFUSE_NS_BASE + EFUSE_T_CSB_R);writel(EFUSE_TIMING(1, 4), EFUSE_NS_BASE + EFUSE_T_PGENB_R);writel(EFUSE_TIMING(1, 4), EFUSE_NS_BASE + EFUSE_T_LOAD_R);writel(EFUSE_TIMING(1, 4), EFUSE_NS_BASE + EFUSE_T_ADDR_R);writel(EFUSE_TIMING(2, 3), EFUSE_NS_BASE + EFUSE_T_STROBE_R);Shouldn't this be done in the efuse driver instead?
I was considering the same and was unsure if these values would need to be initialized at EL3 in SPL or could be initialized at EL1 in U-Boot proper (in the efuse driver).
Did not have time to investigate further and decided to just do it at EL3 same as vendor TF-A, so code ended up here.
Is there a reason not to do it in the upstream TF-A like the vendor one does? I'm sure I've seen patches around that do this when I've dug into it in the past. Is there reasons where TF-A would need access to the information the efuses provide, I'm guessing things like serial numbers etc, given the vendor TF-A does this I suspect that's where the patch should go rather than U-Boot.
Peter