Re: [PATCH v5 5/5] msm: add SMP support for msm

[Catalin Marinas <catalin.marinas@arm.com>]

On 14 December 2010 04:50, Jeff Ohlstein <johlstei@codeaurora.org> wrote:
> --- /dev/null
> +++ b/arch/arm/mach-msm/headsmp.S
[...]
> +ENTRY(msm_secondary_startup)
> +       mrc     p15, 0, r0, c0, c0, 5   @ MPIDR
> +       and     r0, r0, #15             @ What CPU am I
> +       adr     r4, 1f                  @ address of
> +       ldmia   r4, {r5, r6}            @ load curr addr and pen_rel addr
> +       sub     r4, r4, r5              @ determine virtual/phys offsets
> +       add     r6, r6, r4              @ apply
> +pen:
> +       wfe
> +       dsb                             @ ensure subsequent access is
> +                                       @ after event

As I wrote in a different e-mail, this DSB isn't actually needed as
the processor shouldn't speculate beyond the WFE. But it's harmless
anyway.

> --- /dev/null
> +++ b/arch/arm/mach-msm/platsmp.c
[...]
> +int boot_secondary(unsigned int cpu, struct task_struct *idle)
> +{
> +       static int cold_boot_done;
> +       unsigned long timeout;
> +       printk(KERN_DEBUG "Starting secondary CPU %d\n", cpu);
> +
> +       if (cold_boot_done == false) {
> +               prepare_cold_cpu(cpu);
> +               cold_boot_done = true;
> +       }
> +
> +       pen_release = cpu;
> +       __cpuc_flush_dcache_area((void *)&pen_release, sizeof(pen_release));
> +       outer_clean_range(__pa(&pen_release), __pa(&pen_release + 1));
> +       __asm__("sev");
> +       dsb();

What's the need for the dsb() here (one before SEV maybe)?

Also, do we need to add a "memory" clobber to the SEV asm, not sure
whether the compiler would try to reorder the code.

-- 
Catalin