ARM: shmobile: sh73a0: CPU1 failed to come online (Re: [PATCH 0/4] ARM: shmobile: r8a7740/armadillo:

ARM: shmobile: sh73a0: CPU1 failed to come online (Re: [PATCH 0/4] ARM: shmobile: r8a7740/armadillo:

[Geert Uytterhoeven]

On Mon, Nov 17, 2014 at 2:53 PM, Geert Uytterhoeven
<geert+renesas@glider.be> wrote:
> This patch series migrates the DT-based generic r8a7740 platform and the
> armadillo legacy platform from calling l2x0_of_init() to the generic l2c OF
> initialization.
>
> Note that this conversion should be applied to sh73a0, r8a7778, and r8a7779,
> too. More about sh73a0 later.

I tried to convert sh73a0, but it doesn't work, due to problems
bringing up the second
CPU core.

Details:
  - l2x0_of_init() is called much earlier than l2x0_init().
    When the L2 cache is enabled, initialization of the second CPU core of
    sh73a0 (Dual-core Cortex-A9) fails.
  - As jiffies are not yet incrementing at that point (they are on r8a7791?),
    timeouts don't work, and the kernel hangs at the
    wait_for_completion_timeout() call in arch/arm/kernel/smp.c:__cpu_up().
  - This may be related to the intermittent failure to bring up the second
    CPU core when resuming from s2ram:

        CPU1: failed to come online

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

Re: ARM: shmobile: sh73a0: CPU1 failed to come online (Re: [PATCH 0/4] ARM: shmobile: r8a7740/armadi

[Russell King - ARM Linux]

On Mon, Nov 17, 2014 at 02:57:46PM +0100, Geert Uytterhoeven wrote:
> I tried to convert sh73a0, but it doesn't work, due to problems
> bringing up the second
> CPU core.
> 
> Details:
>   - l2x0_of_init() is called much earlier than l2x0_init().
>     When the L2 cache is enabled, initialization of the second CPU core of
>     sh73a0 (Dual-core Cortex-A9) fails.

Missing L2 cache flushes for (any) the secondary core data/trampoline?

>   - As jiffies are not yet incrementing at that point (they are on r8a7791?),
>     timeouts don't work, and the kernel hangs at the
>     wait_for_completion_timeout() call in arch/arm/kernel/smp.c:__cpu_up().

That sounds bad:

start_kernel() -> rest_init() -> kernel_thread(kernel_init)

kernel_init() -> kernel_init_freeable() -> smp_init() -> (for each CPU) cpu_up()

By the time we get to rest_init(), init_IRQ(), time_init(),
local_irq_enable(), late_time_init(), calibrate_delay(), etc have all
been called, so if jiffies is not incrementing in __cpu_up(), that's
a big problem.

-- 
FTTC broadband for 0.8mile line: currently at 9.5Mbps down 400kbps up
according to speedtest.net.

Re: ARM: shmobile: sh73a0: CPU1 failed to come online (Re: [PATCH 0/4] ARM: shmobile: r8a7740/armadi

[Geert Uytterhoeven]

On Mon, Nov 17, 2014 at 4:12 PM, Russell King - ARM Linux
<linux@arm.linux.org.uk> wrote:
> On Mon, Nov 17, 2014 at 02:57:46PM +0100, Geert Uytterhoeven wrote:
>> I tried to convert sh73a0, but it doesn't work, due to problems
>> bringing up the second
>> CPU core.
>>
>> Details:
>>   - l2x0_of_init() is called much earlier than l2x0_init().
>>     When the L2 cache is enabled, initialization of the second CPU core of
>>     sh73a0 (Dual-core Cortex-A9) fails.
>
> Missing L2 cache flushes for (any) the secondary core data/trampoline?

Thanks, I added a few cache flushes, so far it didn't help...

>>   - As jiffies are not yet incrementing at that point (they are on r8a7791?),
>>     timeouts don't work, and the kernel hangs at the
>>     wait_for_completion_timeout() call in arch/arm/kernel/smp.c:__cpu_up().
>
> That sounds bad:
>
> start_kernel() -> rest_init() -> kernel_thread(kernel_init)
>
> kernel_init() -> kernel_init_freeable() -> smp_init() -> (for each CPU) cpu_up()
>
> By the time we get to rest_init(), init_IRQ(), time_init(),
> local_irq_enable(), late_time_init(), calibrate_delay(), etc have all
> been called, so if jiffies is not incrementing in __cpu_up(), that's
> a big problem.

That was due to the missing arm,cortex-a9-twd-timer timer, which also
broke CONFIG_CPU_IDLE=y.

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds