On 7/25/19 12:41 PM, Johann Obermayr via Xenomai wrote:
> Hello,
>
> We are now migrate from Xenomai 2.6.4 + Kernel 3.10.53 to Xenomai 3.x +
> Kernel 4.14.89 (88).
>
> Now I have found follow commit in the old kernel + ipipe
>
> * ipipe/arm: fix system lockup on reboot
>
> This fixes the problem of the system being stuck in an infinite loop at
> LDREX/STREX at the late stage of the reboot sequence.
>
> At this stage the mmu is set into a state so that soft boot works, but causing
> a LDREX/STREX loop to fail. So we have to also disable Xenomai interrupts.
>
> The local_irq_disable() is changed for hard_local_irq_disable() to switch off
> all
> irq's.
>
> arch/arm/kernel/process.c | 13 +++++++++++++
> 1 file changed, 13 insertions(+)
>
> diff --git a/arch/arm/kernel/process.c b/arch/arm/kernel/process.c
> index 0707d004ac79..0e725c76a777 100644
> --- a/arch/arm/kernel/process.c
> +++ b/arch/arm/kernel/process.c
> @@ -98,8 +98,13 @@ void soft_restart(unsigned long addr)
> u64 *stack = soft_restart_stack + ARRAY_SIZE(soft_restart_stack);
> /* Disable interrupts first */
> +#ifdef CONFIG_IPIPE
> + hard_local_irq_disable();
> + hard_local_fiq_disable_notrace();
> +#else
> local_irq_disable();
> local_fiq_disable();
> +#endif
> /* Disable the L2 if we're the last man standing. */
> if (num_online_cpus() == 1)
> @@ -241,7 +246,11 @@ void machine_halt(void)
> {
> smp_send_stop();
> +#ifdef CONFIG_IPIPE
> + hard_local_irq_disable();
> +#else
> local_irq_disable();
> +#endif
> while (1);
> }
> @@ -281,7 +290,11 @@ void machine_restart(char *cmd)
> /* Whoops - the platform was unable to reboot. Tell the user! */
> printk("Reboot failed -- System halted\n");
> +#ifdef CONFIG_IPIPE
> + hard_local_irq_disable();
> +#else
> local_irq_disable();
> +#endif
> while (1);
> }
>
>
> And i don't known, if this patch is also needed for the new version.
>
It does, but now this code lives in arch/arm/kernel/reboot.c.
local_irq_disable() can be replaced by hard_local_irq_disable() without
any conditional ifdeffery. raw_local_irq_disable() is a different beast
that does not map exactely to the hard_* API, so you still need
conditional blocks to set hard_local_irq_disable() when pipelining.
--
Philippe.
