[RFC/PATCH] ARM: vDSO gettimeofday using generic timer architecture

[Russell King - ARM Linux]

On Tue, Jan 28, 2014 at 03:06:53PM -0600, Nathan Lynch wrote:
> +static union {
> +	struct vdso_data	data;
> +	u8			page[PAGE_SIZE];
> +} vdso_data_store __page_aligned_data;
> +struct vdso_data *vdso_data = &vdso_data_store.data;

So this is in the kernel data segment.

> +void update_vsyscall(struct timekeeper *tk)
> +{
> +	struct timespec xtime_coarse;
> +	struct timespec wall_time = tk_xtime(tk);
> +	struct timespec *wtm = &tk->wall_to_monotonic;
> +	u32 use_syscall = strcmp(tk->clock->name, "arch_sys_counter");
> +
> +	++vdso_data->tb_seq_count;
> +	smp_wmb();
> +
> +	xtime_coarse = __current_kernel_time();
> +	vdso_data->use_syscall			= use_syscall;
> +	vdso_data->xtime_coarse_sec		= xtime_coarse.tv_sec;
> +	vdso_data->xtime_coarse_nsec		= xtime_coarse.tv_nsec;
> +
> +	if (!use_syscall) {
> +		vdso_data->cs_cycle_last	= tk->clock->cycle_last;
> +		vdso_data->xtime_clock_sec	= wall_time.tv_sec;
> +		vdso_data->xtime_clock_nsec	= wall_time.tv_nsec;
> +		vdso_data->cs_mult		= tk->mult;
> +		vdso_data->cs_shift		= tk->shift;
> +		vdso_data->wtm_clock_sec	= wtm->tv_sec;
> +		vdso_data->wtm_clock_nsec	= wtm->tv_nsec;
> +	}
> +
> +	smp_wmb();
> +	++vdso_data->tb_seq_count;
> +}
> +
> +void update_vsyscall_tz(void)
> +{
> +	vdso_data->tz_minuteswest	= sys_tz.tz_minuteswest;
> +	vdso_data->tz_dsttime		= sys_tz.tz_dsttime;
> +}

which gets written to directly, and read from userspace.  This won't work
with aliasing caches, of which we have VIVT caches on all ARMv4 and ARMv5
CPUs, and VIPT caches on some ARMv6.

Either this needs to be limited to just VIPT nonaliasing caches, or it
needs cache handling.

The above also looks rather unsafe from the SMP perspective - how does
vdso_data->tb_seq_count protect this data between CPUs?

-- 
FTTC broadband for 0.8mile line: 5.8Mbps down 500kbps up.  Estimation
in database were 13.1 to 19Mbit for a good line, about 7.5+ for a bad.
Estimate before purchase was "up to 13.2Mbit".