Re: [PATCH v4 1/2] clk: rockchip: rk3288: Handle clock tree for rk3288w

From: Robin Murphy
Date: Fri Jul 03 2020 - 11:43:24 EST


On 2020-07-03 15:48, Heiko Stuebner wrote:
Am Freitag, 3. Juli 2020, 16:23:27 CEST schrieb Jagan Teki:
On Fri, Jul 3, 2020 at 7:41 PM Heiko Stuebner <heiko@xxxxxxxxx> wrote:

Hi Jagan,

Am Montag, 29. Juni 2020, 21:11:03 CEST schrieb Jagan Teki:
On Tue, Jun 2, 2020 at 1:37 PM MylÃne Josserand
<mylene.josserand@xxxxxxxxxxxxx> wrote:

The revision rk3288w has a different clock tree about "hclk_vio"
clock, according to the BSP kernel code.

This patch handles this difference by detecting which device-tree
we are using. If it is a "rockchip,rk3288-cru", let's register
the clock tree as it was before. If the device-tree node is
"rockchip,rk3288w-cru", we will apply the difference with this
version of this SoC.

Noticed that this new device-tree compatible must be handled in
bootloader such as u-boot.

Signed-off-by: MylÃne Josserand <mylene.josserand@xxxxxxxxxxxxx>
---
drivers/clk/rockchip/clk-rk3288.c | 20 ++++++++++++++++++--
1 file changed, 18 insertions(+), 2 deletions(-)

diff --git a/drivers/clk/rockchip/clk-rk3288.c b/drivers/clk/rockchip/clk-rk3288.c
index cc2a177bbdbf..204976e2d0cb 100644
--- a/drivers/clk/rockchip/clk-rk3288.c
+++ b/drivers/clk/rockchip/clk-rk3288.c
@@ -425,8 +425,6 @@ static struct rockchip_clk_branch rk3288_clk_branches[] __initdata = {
COMPOSITE(0, "aclk_vio0", mux_pll_src_cpll_gpll_usb480m_p, CLK_IGNORE_UNUSED,
RK3288_CLKSEL_CON(31), 6, 2, MFLAGS, 0, 5, DFLAGS,
RK3288_CLKGATE_CON(3), 0, GFLAGS),
- DIV(0, "hclk_vio", "aclk_vio0", 0,
- RK3288_CLKSEL_CON(28), 8, 5, DFLAGS),
COMPOSITE(0, "aclk_vio1", mux_pll_src_cpll_gpll_usb480m_p, CLK_IGNORE_UNUSED,
RK3288_CLKSEL_CON(31), 14, 2, MFLAGS, 8, 5, DFLAGS,
RK3288_CLKGATE_CON(3), 2, GFLAGS),
@@ -819,6 +817,16 @@ static struct rockchip_clk_branch rk3288_clk_branches[] __initdata = {
INVERTER(0, "pclk_isp", "pclk_isp_in", RK3288_CLKSEL_CON(29), 3, IFLAGS),
};

+static struct rockchip_clk_branch rk3288w_hclkvio_branch[] __initdata = {
+ DIV(0, "hclk_vio", "aclk_vio1", 0,
+ RK3288_CLKSEL_CON(28), 8, 5, DFLAGS),
+};
+
+static struct rockchip_clk_branch rk3288_hclkvio_branch[] __initdata = {
+ DIV(0, "hclk_vio", "aclk_vio0", 0,
+ RK3288_CLKSEL_CON(28), 8, 5, DFLAGS),
+};
+
static const char *const rk3288_critical_clocks[] __initconst = {
"aclk_cpu",
"aclk_peri",
@@ -936,6 +944,14 @@ static void __init rk3288_clk_init(struct device_node *np)
RK3288_GRF_SOC_STATUS1);
rockchip_clk_register_branches(ctx, rk3288_clk_branches,
ARRAY_SIZE(rk3288_clk_branches));
+
+ if (of_device_is_compatible(np, "rockchip,rk3288w-cru"))
+ rockchip_clk_register_branches(ctx, rk3288w_hclkvio_branch,
+ ARRAY_SIZE(rk3288w_hclkvio_branch));
+ else
+ rockchip_clk_register_branches(ctx, rk3288_hclkvio_branch,
+ ARRAY_SIZE(rk3288_hclkvio_branch));
+

Sorry for the late query on this. I am a bit unclear about this
compatible change, does Linux expect to replace rockchip,rk3288-cru
with rockchip,rk3288w-cru in bootloader if the chip is RK3288w? or
append the existing cru compatible node with rockchip,rk3288w-cru?
because replace new cru node make clock never probe since the
CLK_OF_DECLARE checking rockchip,rk3288-cru

I guess right now we'd expect "rockchip,rk3288w-cru", "rockchip,rk3288-cru",

Thinking again about this, I'm wondering if we should switch to having
only one per variant ... like on the two rk3188 variants,
so declaring separate rk3288-cru and rk3288w-cru of-clks with shared
common code.

What if can check the root compatible instead cru compatible for revision W like

- if (of_device_is_compatible(np, "rockchip,rk3288w-cru"))
+ if (of_device_is_compatible(np, "rockchip,rk3288w"))

you'd need to check against the root compatible.


This way we can have a single compatible update at bootloader that
makes Linux adjust revision W chips code.

Doesn't it make sense?

The compatible describes the block and the rk3288w's cru isn't the same as
the rk3288's ... as the clock routing is different, so it should have a
different compatible value, I think.

Right, if two devices behave differently in a way that one behaviour is not a strict superset of the other, and there's no way to tell which is which by simply looking at the device itself, then by definition they are not compatible.

As the DT sis supposed to be a _generic_ description of the hardware,
we don't want to cement hacks to other implementations would need to copy.

Indeed it's a pain in the bum for driver developers to be given a new thing that's 99.9% compatible with the old thing but the one tiny difference fundamentally breaks it, and the temptation to look for an easy way out is strong, but c'est la vie ;)

The question I always bear in mind for cases like these is this: If a kernel without the driver patch runs on the new system with the expected new DTB, will it:

A) work entirely correctly
B) ignore the device (with or without an error), but the rest of the system still works fine
C) ignore the device and fail to boot at all (with or without an error) because it's critical to operation of the rest of the system
D) probe the device and superficially appear to be OK but actually be broken in weird and subtle ways

A and B are the ideal answers. C is a little unfortunate, but acceptable if neither A nor B is possible. D is a sign that you're doing something wrong, unless there are *very* specific circumstances that might justify it (e.g. the SoC is exclusively deployed in embedded devices with a controlled update mechanism such that users could never introduce this mismatch in the first place).

Robin.