arm64: vipi: vIPI code adjustments

Some systems, such as the Apple M1 SoC, do not have more than one or
two hardware IPIs that can be distinguished and masked
independently. For such systems, it is necessary to provide "virtual"
IPIs, but since there are at least two IRQ controllers that need this
feature, let's move it into common code.

The new entrypoint accepts a potentially non-maskable single per-CPU
IPI interrupt and provides up to 32 "virtual" per-CPU IPIs that are
maskable, can be distinguished, and live in an IPI domain.

set_smp_ipi_range is changed to transparently enable the vIPI layer on
IPI-challenged systems.

TODO: move vipi_init definition to header file
TODO: return a value from set_smp_ipi_range so drivers know whether to
set irq domain flags.
TODO: make conditional on IPI-challenged archs.

Signed-off-by: Pip Cet <pipcet@gmail.com>

arch/arm64/Kconfig

@@ -314,6 +314,9 @@ config ARCH_MHP_MEMMAP_ON_MEMORY_ENABLE
 config SMP
 	def_bool y
 
+config IPI_FUNNELING
+        def_bool y
+
 config KERNEL_MODE_NEON
 	def_bool y
 

arch/arm64/kernel/Makefile

@@ -73,6 +73,7 @@ obj-$(CONFIG_ARM64_PTR_AUTH)		+= pointer_auth.o
 obj-$(CONFIG_ARM64_MTE)			+= mte.o
 obj-y					+= vdso-wrap.o
 obj-$(CONFIG_COMPAT_VDSO)		+= vdso32-wrap.o
+obj-$(CONFIG_IPI_FUNNELING)		+= vipi.o
 
 obj-y					+= probes/
 head-y					:= head.o

arch/arm64/kernel/smp.c

@@ -979,10 +979,26 @@ static void ipi_teardown(int cpu)
 }
 #endif
 
+#ifdef CONFIG_IPI_FUNNELING
+extern int __init vipi_init(struct irq_data *hwirq);
+#else
+static inline int vipi_init(irqdata)
+{
+	return -EINVAL;
+}
+#endif
+
 void __init set_smp_ipi_range(int ipi_base, int n)
 {
 	int i;
 
+	if (n < NR_IPI) {
+		int ret;
+		BUG_ON(n < 1);
+		ret = vipi_init(irq_get_irq_data(ipi_base));
+		if (ret >= 0)
+			return;
+	}
 	WARN_ON(n < NR_IPI);
 	nr_ipi = min(n, NR_IPI);
 

arch/arm64/kernel/vipi.c

@@ -21,6 +21,19 @@
 #include <asm/sysreg.h>
 #include <asm/virt.h>
 
+struct vipi_irq_chip {
+	struct irq_domain *domain;
+	struct irq_data *hwirq;
+};
+
+#define NR_SWIPI 32
+
+static DEFINE_PER_CPU(atomic_t, vipi_flag);
+static DEFINE_PER_CPU(atomic_t, vipi_enable);
+
+static struct vipi_irq_chip *vipi_irqc;
+
+static void handle_ipi(struct irq_desc *d);
 /*
  * IPI irqchip
  */
@@ -83,8 +96,10 @@ static void vipi_send_mask(struct irq_data *d, const struct cpumask *mask)
 		smp_mb__after_atomic();
 
 		if (!(pending & irq_bit) &&
-		    (atomic_read(per_cpu_ptr(&aic_vipi_enable, cpu)) & irq_bit))
-			send |= AIC_IPI_SEND_CPU(cpu);
+		    (atomic_read(per_cpu_ptr(&vipi_enable, cpu)) & irq_bit)) {
+			cpumask_set_cpu(cpu, &sendmask);
+			send = true;
+		}
 	}
 
 	/*
@@ -94,7 +109,7 @@ static void vipi_send_mask(struct irq_data *d, const struct cpumask *mask)
 	 * already ordered after the vIPI flag write.
 	 */
 	if (send)
-		aic_ic_write(ic, AIC_IPI_SEND, send);
+		ipi_send_mask(ic->hwirq->irq, &sendmask);
 }
 
 static struct irq_chip vipi_chip = {
@@ -132,9 +147,12 @@ static void handle_ipi(struct irq_desc *d)
 	 */
 	firing = atomic_fetch_andnot(enabled, this_cpu_ptr(&vipi_flag)) & enabled;
 
-	for_each_set_bit(i, &firing, AIC_NR_SWIPI)
-		handle_domain_irq(aic_irqc->ipi_domain, i, regs);
+	for_each_set_bit(i, &firing, NR_SWIPI) {
+		struct irq_desc *nd =
+			irq_resolve_mapping(vipi_irqc->domain, i);
 
+		handle_irq_desc(nd);
+	}
 }
 
 static int vipi_alloc(struct irq_domain *d, unsigned int virq,
@@ -154,6 +172,7 @@ static int vipi_alloc(struct irq_domain *d, unsigned int virq,
 static void vipi_free(struct irq_domain *d, unsigned int virq, unsigned int nr_irqs)
 {
 	/* Not freeing IPIs */
+	WARN_ON(1);
 }
 
 static const struct irq_domain_ops vipi_domain_ops = {
@@ -165,8 +184,12 @@ static int vipi_init_smp(struct vipi_irq_chip *irqc)
 {
 	struct irq_domain *vipi_domain;
 	int base_ipi;
+	struct fwnode_handle *fwnode;
+
+	fwnode = __irq_domain_alloc_fwnode(IRQCHIP_FWNODE_NAMED, 0,
+					   "vIPI", NULL);
 
-	ipi_domain = irq_domain_create_linear(irqc->hw_domain->fwnode, AIC_NR_SWIPI,
+	vipi_domain = irq_domain_create_linear(fwnode, NR_SWIPI,
 					      &vipi_domain_ops, irqc);
 	if (WARN_ON(!vipi_domain))
 		return -ENOMEM;
@@ -179,7 +202,7 @@ static int vipi_init_smp(struct vipi_irq_chip *irqc)
 
 	if (WARN_ON(base_ipi < 0)) {
 		irq_domain_remove(vipi_domain);
-		return -ENODEV;
+		return -ENOMEM;
 	}
 
 	set_smp_ipi_range(base_ipi, NR_SWIPI);
@@ -189,19 +212,24 @@ static int vipi_init_smp(struct vipi_irq_chip *irqc)
 	return 0;
 }
 
-static int aic_init_cpu(unsigned int cpu)
+int __init vipi_init(struct irq_data *hwirq)
 {
+	struct vipi_irq_chip *irqc;
 
 	irqc = kzalloc(sizeof(*irqc), GFP_KERNEL);
 	if (!irqc)
 		return -ENOMEM;
 
+	irqc->hwirq = hwirq;
 
+	if (vipi_init_smp(irqc))
+		return -ENOMEM;
 
+	vipi_irqc = irqc;
 
+	irq_set_handler_locked(hwirq, handle_ipi);
 
-	pr_info("Initialized with %d IRQs, %d FIQs, %d vIPIs\n",
-		irqc->nr_hw, AIC_NR_FIQ, AIC_NR_SWIPI);
+	pr_info("Initialized with %d vIPIs\n", NR_SWIPI);
 
 	return 0;
 }