On the A31, the DMA engine only works if AHB1 is clocked from PLL6.
In addition, the hstimer is clocked from AHB1, and if AHB1 is clocked
from the CPU clock, and cpufreq is working, we get an unstable timer.
Force the AHB1 clock to use PLL6 as its parent. Previously this was done
in the device tree with the assigned-clocks and assigned-clocks-parent
bindings. However with this new monolithic driver, the system critical
clocks aren't exported through the device tree. The alternative is to
force this setting in the driver before the clocks are registered.
This is also done in newer versions of mainline U-boot. But people still
using an older version, or even the vendor version, can still hit this
issue. Hence the need to do it in the kernel as well.
Reported-by: Hans de Goede <[email protected]>
Reported-by: Maxime Ripard <[email protected]>
Fixes: c6e6c96d8fa6 ("clk: sunxi-ng: Add A31/A31s clocks")
Signed-off-by: Chen-Yu Tsai <[email protected]>
---
Hi,
This fix is for 4.9. This should fix issues with hstimer and DMA when
booting with an old bootloader.
ChenYu
---
drivers/clk/sunxi-ng/ccu-sun6i-a31.c | 12 ++++++++++++
1 file changed, 12 insertions(+)
diff --git a/drivers/clk/sunxi-ng/ccu-sun6i-a31.c b/drivers/clk/sunxi-ng/ccu-sun6i-a31.c
index 79596463e0d9..4a82a49cff5e 100644
--- a/drivers/clk/sunxi-ng/ccu-sun6i-a31.c
+++ b/drivers/clk/sunxi-ng/ccu-sun6i-a31.c
@@ -191,6 +191,8 @@ static struct clk_div_table axi_div_table[] = {
static SUNXI_CCU_DIV_TABLE(axi_clk, "axi", "cpu",
0x050, 0, 3, axi_div_table, 0);
+#define SUN6I_A31_AHB1_REG 0x054
+
static const char * const ahb1_parents[] = { "osc32k", "osc24M",
"axi", "pll-periph" };
@@ -1230,6 +1232,16 @@ static void __init sun6i_a31_ccu_setup(struct device_node *node)
val &= BIT(16);
writel(val, reg + SUN6I_A31_PLL_MIPI_REG);
+ /* Force AHB1 to PLL6 / 3 */
+ val = readl(reg + SUN6I_A31_AHB1_REG);
+ /* set PLL6 pre-div = 3 */
+ val &= ~GENMASK(7, 6);
+ val |= 0x2 << 6;
+ /* select PLL6 / pre-div */
+ val &= ~GENMASK(13, 12);
+ val |= 0x3 << 12;
+ writel(val, reg + SUN6I_A31_AHB1_REG);
+
sunxi_ccu_probe(node, reg, &sun6i_a31_ccu_desc);
ccu_mux_notifier_register(pll_cpu_clk.common.hw.clk,
--
2.9.3
On Tue, 18 Oct 2016 13:42:09 +0800
Chen-Yu Tsai <[email protected]> wrote:
> On the A31, the DMA engine only works if AHB1 is clocked from PLL6.
> In addition, the hstimer is clocked from AHB1, and if AHB1 is clocked
> from the CPU clock, and cpufreq is working, we get an unstable timer.
>
> Force the AHB1 clock to use PLL6 as its parent. Previously this was done
> in the device tree with the assigned-clocks and assigned-clocks-parent
> bindings. However with this new monolithic driver, the system critical
> clocks aren't exported through the device tree. The alternative is to
> force this setting in the driver before the clocks are registered.
It should be simpler to export the constant (CLK_AHB1) instead of
adding code...
--
Ken ar c'henta? | ** Breizh ha Linux atav! **
Jef | http://moinejf.free.fr/
On Tue, Oct 18, 2016 at 3:50 PM, Jean-Francois Moine <[email protected]> wrote:
> On Tue, 18 Oct 2016 13:42:09 +0800
> Chen-Yu Tsai <[email protected]> wrote:
>
>> On the A31, the DMA engine only works if AHB1 is clocked from PLL6.
>> In addition, the hstimer is clocked from AHB1, and if AHB1 is clocked
>> from the CPU clock, and cpufreq is working, we get an unstable timer.
>>
>> Force the AHB1 clock to use PLL6 as its parent. Previously this was done
>> in the device tree with the assigned-clocks and assigned-clocks-parent
>> bindings. However with this new monolithic driver, the system critical
>> clocks aren't exported through the device tree. The alternative is to
>> force this setting in the driver before the clocks are registered.
>
> It should be simpler to export the constant (CLK_AHB1) instead of
> adding code...
I get you. But the plan is to not export system core clocks that
don't have a direct user.
ChenYu
On Tue, Oct 18, 2016 at 01:42:09PM +0800, Chen-Yu Tsai wrote:
> On the A31, the DMA engine only works if AHB1 is clocked from PLL6.
> In addition, the hstimer is clocked from AHB1, and if AHB1 is clocked
> from the CPU clock, and cpufreq is working, we get an unstable timer.
>
> Force the AHB1 clock to use PLL6 as its parent. Previously this was done
> in the device tree with the assigned-clocks and assigned-clocks-parent
> bindings. However with this new monolithic driver, the system critical
> clocks aren't exported through the device tree. The alternative is to
> force this setting in the driver before the clocks are registered.
>
> This is also done in newer versions of mainline U-boot. But people still
> using an older version, or even the vendor version, can still hit this
> issue. Hence the need to do it in the kernel as well.
>
> Reported-by: Hans de Goede <[email protected]>
> Reported-by: Maxime Ripard <[email protected]>
> Fixes: c6e6c96d8fa6 ("clk: sunxi-ng: Add A31/A31s clocks")
> Signed-off-by: Chen-Yu Tsai <[email protected]>
Applied, thanks!
Maxime
--
Maxime Ripard, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com