port-arm/58593: tegra_cpufreq lacks memory barriers in bespoke locking

[campbell+netbsd%mumble.net@localhost]

>Number:         58593
>Category:       port-arm
>Synopsis:       tegra_cpufreq lacks memory barriers in bespoke locking
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    port-arm-maintainer
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Wed Aug 14 02:00:00 +0000 2024
>Originator:     Taylor R Campbell
>Release:        current, 10
>Organization:
The NvidiaBSD Frequentegra
>Environment:
>Description:
    153 	if (atomic_cas_uint(&cpufreq_busy, 0, 1) != 0)
    154 		return EBUSY;
    155 
    156 	error = cpufreq_set_rate(fq);
    157 	if (error == 0) {
    158 		pmf_event_inject(NULL, PMFE_SPEED_CHANGED);
    159 	}
    160 
    161 	atomic_dec_uint(&cpufreq_busy);

https://nxr.netbsd.org/xref/src/sys/arch/arm/nvidia/tegra_cpufreq.c?r=1.5#153

This doesn't provide adequate memory ordering.
>How-To-Repeat:
code inspection
>Fix:
Two options:

1. After atomic_cas_uint, add membar_acquire; and add membar_release before atomic_dec_uint.

2. Just use mutex(9).