[PATCH 20/22] ARM: sun8i: Add SMP support for the Allwinner A23

[Marc Zyngier]

On 26/05/14 04:57, Chen-Yu Tsai wrote:
> On Mon, May 26, 2014 at 3:26 AM, Maxime Ripard
> <maxime.ripard at free-electrons.com> wrote:
>> On Fri, May 23, 2014 at 03:51:23PM +0800, Chen-Yu Tsai wrote:
>>> The A23 is a dual Cortex-A7. Add the logic to use the IPs used to
>>> control the CPU configuration and the CPU power so that we can
>>> bring up secondary CPUs at boot.
>>>
>>> Signed-off-by: Chen-Yu Tsai <wens at csie.org>
>>> ---
>>>  arch/arm/mach-sunxi/platsmp.c | 69 +++++++++++++++++++++++++++++++++++++++++++
>>>  1 file changed, 69 insertions(+)
>>>
>>> diff --git a/arch/arm/mach-sunxi/platsmp.c b/arch/arm/mach-sunxi/platsmp.c
>>> index c53077b..688faaf 100644
>>> --- a/arch/arm/mach-sunxi/platsmp.c
>>> +++ b/arch/arm/mach-sunxi/platsmp.c
>>> @@ -121,3 +121,72 @@ struct smp_operations sun6i_smp_ops __initdata = {
>>>       .smp_boot_secondary     = sun6i_smp_boot_secondary,
>>>  };
>>>  CPU_METHOD_OF_DECLARE(sun6i_smp, "allwinner,sun6i-a31", &sun6i_smp_ops);
>>> +
>>> +static void __init sun8i_smp_prepare_cpus(unsigned int max_cpus)
>>> +{
>>> +     struct device_node *node;
>>> +
>>> +     node = of_find_compatible_node(NULL, NULL, "allwinner,sun8i-a23-prcm");
>>> +     if (!node) {
>>> +             pr_err("Missing A23 PRCM node in the device tree\n");
>>> +             return;
>>> +     }
>>> +
>>> +     prcm_membase = of_iomap(node, 0);
>>> +     if (!prcm_membase) {
>>> +             pr_err("Couldn't map A23 PRCM registers\n");
>>> +             return;
>>> +     }
>>> +
>>> +     node = of_find_compatible_node(NULL, NULL,
>>> +                                    "allwinner,sun8i-a23-cpuconfig");
>>> +     if (!node) {
>>> +             pr_err("Missing A23 CPU config node in the device tree\n");
>>> +             return;
>>> +     }
>>> +
>>> +     cpucfg_membase = of_iomap(node, 0);
>>> +     if (!cpucfg_membase)
>>> +             pr_err("Couldn't map A23 CPU config registers\n");
>>> +
>>> +}
>>> +
>>> +static int sun8i_smp_boot_secondary(unsigned int cpu,
>>> +                                 struct task_struct *idle)
>>> +{
>>> +     u32 reg;
>>> +
>>> +     if (!(prcm_membase && cpucfg_membase))
>>> +             return -EFAULT;
>>> +
>>> +     spin_lock(&cpu_lock);
>>> +
>>> +     /* Set CPU boot address */
>>> +     writel(virt_to_phys(secondary_startup),
>>> +            cpucfg_membase + CPUCFG_PRIVATE0_REG);
>>> +
>>> +     /* Assert the CPU core in reset */
>>> +     writel(0, cpucfg_membase + CPUCFG_CPU_RST_CTRL_REG(cpu));
>>> +
>>> +     /* Assert the L1 cache in reset */
>>> +     reg = readl(cpucfg_membase + CPUCFG_GEN_CTRL_REG);
>>> +     writel(reg & ~BIT(cpu), cpucfg_membase + CPUCFG_GEN_CTRL_REG);
>>> +
>>> +     /* Clear CPU power-off gating */
>>> +     reg = readl(prcm_membase + PRCM_CPU_PWROFF_REG);
>>> +     writel(reg & ~BIT(cpu), prcm_membase + PRCM_CPU_PWROFF_REG);
>>> +     mdelay(1);
>>> +
>>> +     /* Deassert the CPU core reset */
>>> +     writel(3, cpucfg_membase + CPUCFG_CPU_RST_CTRL_REG(cpu));
>>> +
>>> +     spin_unlock(&cpu_lock);
>>> +
>>> +     return 0;
>>> +}
>>> +
>>> +struct smp_operations sun8i_smp_ops __initdata = {
>>> +     .smp_prepare_cpus       = sun8i_smp_prepare_cpus,
>>> +     .smp_boot_secondary     = sun8i_smp_boot_secondary,
>>> +};
>>> +CPU_METHOD_OF_DECLARE(sun8i_smp, "allwinner,sun8i-a23", &sun8i_smp_ops);
>>
>> You forgot to document the new enable-method.
> 
> I will add it.
> 
>> Also, is there any plan to hae a working u-boot? I'd much prefer to
>> use PSCI if possible.
> 
> IIRC PSCI needs a secure SRAM block to store its program code.
> Unfortunately the A23 doesn't have secure SRAM. I think it is
> missing other security related features as well.
> 
> Or could it just live in generic SRAM, and the kernel marks it
> as reserved or something.
> 
> Maybe Marc Zyngier (CCed) can shed some light on this?

No secure SRAM is required. You can normal SRAM, or even a normal region
of RAM that is accessed from secure mode. Not as nice as having proper
SRAM as on sun7i, but still perfectly functional.

Please consider having a proper PSCI implementation instead of always
adding more code that will effectively lock people out of using the
virtualization features the core has.

Thanks,

	M.
-- 
Jazz is not dead. It just smells funny...