SMP多核启动

在 Linux系统中,对于多核的ARM芯片而言,在Biotron代码中,每个CPU都会识别自身ID,如果ID是0,则引导Bootloader和 Linux内核执行,如果ID不是0,则Biotron一般在上电时将自身置于WFI或者WFE状态,并等待CPU0给其发CPU核间中断或事件(一般通过SEV指令)以唤醒它。一个典型的多核 Linux启动过程如图20.6所示。

被CPU0唤醒的CPUn可以在运行过程中进行热插拔,譬如运行如下命令即可卸载CPU1,并且将CPUI上的任务全部迁移到其他CPU中:

# echo 0 > /sys/devices/system/cpu/cpu1/online

同理，运行如下命令可以再次启动CPU1：

# echo 1 > /sys/devices/system/cpu/cpu1/online

之后CPU1会主动参与系统中各个CPU之间的运行任务的负载均衡工作；

CPUO唤醒其他CPU的动作在内核中被封装为一个 smp_operations的结构体,对于ARM而言,它定义于 arch/arm/include/asm/smp.h中。该结构体的成员函数如代码清单所示。

struct smp_operations {
#ifdef CONFIG_SMP
	/*
	 * Setup the set of possible CPUs (via set_cpu_possible)
	 */
	void (*smp_init_cpus)(void);
	/*
	 * Initialize cpu_possible map, and enable coherency
	 */
	void (*smp_prepare_cpus)(unsigned int max_cpus);

	/*
	 * Perform platform specific initialisation of the specified CPU.
	 */
	void (*smp_secondary_init)(unsigned int cpu);
	/*
	 * Boot a secondary CPU, and assign it the specified idle task.
	 * This also gives us the initial stack to use for this CPU.
	 */
	int  (*smp_boot_secondary)(unsigned int cpu, struct task_struct *idle);
#ifdef CONFIG_HOTPLUG_CPU
	int  (*cpu_kill)(unsigned int cpu);
	void (*cpu_die)(unsigned int cpu);
	int  (*cpu_disable)(unsigned int cpu);
#endif
#endif
};

CPUO唤醒其他CPU的动作在内核中被封装为一个 smp_operations 的结构体,对于ARM而言,它定义于 arch/arm/include/asm/smp.h中。该结构体的成员函数如代码清单所示。

DT_MACHINE_START(VEXPRESS DT,"ARM-Versatile Express)
.dt_compat = v2m_dt_match,
.smp = smp_ops(express_smp_ops),
.map_io = v2m_dt_map_io,
MACHINE_END

通过 arch/arm/mach-vexpress/platsmp.c的实现代码可以看出, smp_operations的成员函数smp_init_cpus(),即 vexpress_smp_init_cpus调用的ct_ca9x4_init_cpu_map(会探测SoC内CPU核的个数,并通过 set_cpu_possible设置这些CPU可见。

而 smp_operations的成员函数 smp_prepare_cpus,即 vexpress_smp_prepare_cpus则会通过v2m_flags_set( virt_to_phys( versatile_secondary_startup)设置其他CPU的启动地址为versatile_secondary_startup,如代码清单所示。

在smp_prepare_cpus()设置CPU1...n启动地址：

static void __init vexpress_smp_prepare_cpus(unsigned int max_cpus)
{
	/*
	 * Initialise the present map, which describes the set of CPUs
	 * actually populated at the present time.
	 */
	if (ct_desc)
		ct_desc->smp_enable(max_cpus);
	else
		vexpress_dt_smp_prepare_cpus(max_cpus);

	/*
	 * Write the address of secondary startup into the
	 * system-wide flags register. The boot monitor waits
	 * until it receives a soft interrupt, and then the
	 * secondary CPU branches to this address.
	 */
	vexpress_flags_set(virt_to_phys(versatile_secondary_startup));
}

注意这部分具体实现方式是与SOC相关的，由芯片设计及芯片内部的Bootrom决定。对于VEXPRESS来讲，设置方法如下：

void __init v2m_flags_set(u32 data)
{
    writel(~0, v2m_sysreg_base + V2M_SYS_FLAGSCLR);
    writel(data, v2m_sysreg_base + V2M_SYS_FLAGSCLR);
}

即填充 v2m_sysreg_base+V2M_SYS_FLAGSCLR标记清除寄存器为0xFFFFFFFF,将CPU1...n初始启动执行的指令地址填入v2m_sysreg_base+V2M_SYS_FLAGSSET寄存器。这两个地址由芯片内部的Bootrom程序设定的。填入的CPU1...n的起始地址都通过virt_to_phys转化为物理地址，因为此时CPU1...n的MMU尚未开启；

比较关键的是smp_operations的成员函数smp_boot_secondary()，它是完成CPU最终唤醒的工作，对于本例而言，versatile_boot_secondary()；

CPU0通过终端唤醒其他CPU：

/*
 * Write pen_release in a way that is guaranteed to be visible to all
 * observers, irrespective of whether they're taking part in coherency
 * or not.  This is necessary for the hotplug code to work reliably.
 */
static void __cpuinit write_pen_release(int val)
{
	pen_release = val;
	smp_wmb();
	__cpuc_flush_dcache_area((void *)&pen_release, sizeof(pen_release));
	outer_clean_range(__pa(&pen_release), __pa(&pen_release + 1));
}

int __cpuinit versatile_boot_secondary(unsigned int cpu, struct task_struct *idle)
{
	unsigned long timeout;

	/*
	 * Set synchronisation state between this boot processor
	 * and the secondary one
	 */
	spin_lock(&boot_lock);

	/*
	 * This is really belt and braces; we hold unintended secondary
	 * CPUs in the holding pen until we're ready for them.  However,
	 * since we haven't sent them a soft interrupt, they shouldn't
	 * be there.
	 */
	write_pen_release(cpu_logical_map(cpu));

	/*
	 * Send the secondary CPU a soft interrupt, thereby causing
	 * the boot monitor to read the system wide flags register,
	 * and branch to the address found there.
	 */
	arch_send_wakeup_ipi_mask(cpumask_of(cpu));

	timeout = jiffies + (1 * HZ);
	while (time_before(jiffies, timeout)) {
		smp_rmb();
		if (pen_release == -1)
			break;

		udelay(10);
	}

	/*
	 * now the secondary core is starting up let it run its
	 * calibrations, then wait for it to finish
	 */
	spin_unlock(&boot_lock);

	return pen_release != -1 ? -ENOSYS : 0;
}

调用的 write_pen_release会将 pen_release变量设置为要唤醒的CPU核的CPU号 cpu_logical_map(cpu),而后通过 arch_send_wakeup_ipi mask给要唤醒的CPU发IPI中断,这个时候,被唤醒的CPU会退出WFI状态并从前面 smp_operations中的smp_prepare_cpus成员函数,即 vexpress_smp_prepare_cpus里通过 v2m_flags_set()设置的起始地址 versatile_secondary_startup开始执行,如果顺利的话,该CPU会将原先为正数的pen_release写为-1，以便CPU0从等待pen_release成为-1的循环跳出；

versatile_secondary_startup实现于arch/arm/plat-versatile/headsmp.S中，是一段汇编，如下代码所示：

/*
 * Realview/Versatile Express specific entry point for secondary CPUs.
 * This provides a "holding pen" into which all secondary cores are held
 * until we're ready for them to initialise.
 */
ENTRY(versatile_secondary_startup)
	mrc	p15, 0, r0, c0, c0, 5
	bic	r0, #0xff000000
	adr	r4, 1f
	ldmia	r4, {r5, r6}
	sub	r4, r4, r5
	add	r6, r6, r4
pen:	ldr	r7, [r6]
	cmp	r7, r0
	bne	pen

	/*
	 * we've been released from the holding pen: secondary_stack
	 * should now contain the SVC stack for this core
	 */
	b	secondary_startup

	.align
1:	.long	.
	.long	pen_release
ENDPROC(versatile_secondary_startup)

上述循环代码的循环是等待pen_release变量称为CPU0设置的cpu_logical_map(cpu)，一般就直接成立了。第16行调用内核通用的secondary_startup()函数，经过一系列初始化（如MMU等），最终新的被唤醒的CPU将调用smp_operations的smp_secondary_init()的成员函数，对于本例为versatile_secondary_init()；

void __cpuinit versatile_secondary_init(unsigned int cpu)
{
	/*
	 * let the primary processor know we're out of the
	 * pen, then head off into the C entry point
	 */
	write_pen_release(-1);

	/*
	 * Synchronise with the boot thread.
	 */
	spin_lock(&boot_lock);
	spin_unlock(&boot_lock);
}

上述代码会将pen_release写为-1，于是CPU0还在执行代码的versatile_boot_secondary()函数中的如下循环就退出了：

timeout = jiffies + (1 * HZ);
while (time_before(jiffies, timeout)) {
	smp_rmb();
	if (pen_release == -1)
		break;

	udelay(10);
}

这样CPU0就知道目标CPU已经被正确地唤醒,此后CPU0和新唤醒的其他CPU各自运行。整个系统在运行过程中会进行实时进程和正常进程的动态负载均衡。

下图总结了前文提到的vexpress_smp_prepare_cpus()、versatile_boot_secondary()、write_pen_release()、versatile_secondary_startup()、versatile_secondary_init()这些函数的执行顺序；