Re: [PATCH v2 17/17] x86/hyperv: handle IO-APIC when running as root

[Wei Liu]

On Thu, Nov 12, 2020 at 05:56:41PM +0100, Vitaly Kuznetsov wrote:
[...]
> > +static unsigned int hv_ioapic_startup_irq(struct irq_data *data)
> > +{
> > +	u16 status;
> > +	struct IO_APIC_route_entry ire;
> > +	u32 vector;
> > +	struct irq_cfg *cfg;
> > +	int ioapic;
> > +	u8 ioapic_pin;
> > +	int ioapic_id;
> > +	int gsi;
> > +	union entry_union eu;
> > +	struct cpumask *affinity;
> > +	int cpu, vcpu;
> > +	struct hv_interrupt_entry entry;
> > +	struct mp_chip_data *mp_data = data->chip_data;
> > +
> > +	gsi = data->irq;
> > +	cfg = irqd_cfg(data);
> > +	affinity = irq_data_get_effective_affinity_mask(data);
> > +	cpu = cpumask_first_and(affinity, cpu_online_mask);
> > +	vcpu = hv_cpu_number_to_vp_number(cpu);
> > +
> > +	vector = cfg->vector;
> > +
> > +	ioapic = mp_find_ioapic(gsi);
> > +	ioapic_pin = mp_find_ioapic_pin(ioapic, gsi);
> > +	ioapic_id = mpc_ioapic_id(ioapic);
> > +	ire = ioapic_read_entry(ioapic, ioapic_pin);
> > +
> > +	/*
> > +	 * Always try unmapping. We do not have visibility into which whether
> > +	 * an IO-APIC has been mapped or not. We can't use chip_data because it
> > +	 * already points to mp_data.
> > +	 *
> > +	 * We don't use retarget interrupt hypercalls here because Hyper-V
> > +	 * doens't allow root to change the vector or specify VPs outside of
> > +	 * the set that is initially used during mapping.
> > +	 */
> > +	status = hv_unmap_ioapic_interrupt(gsi);
> > +
> > +	if (!(status == HV_STATUS_SUCCESS || status == HV_STATUS_INVALID_PARAMETER)) {
> > +		pr_debug("%s: unexpected unmap status %d\n", __func__, status);
> > +		return -1;
> 
> Nit: the function returns 'unsigned int' but I see other 'irq_startup'
> routines return negative values too, however, they tend to returd
> '-ESOMETHING' so maybe -EFAULT here?
> 

The return type should've been int instead. That's what the function
signature in struct irq_chip looks like.

> > +	}
> > +
> > +	status = hv_map_ioapic_interrupt(ioapic_id, ire.trigger, vcpu, vector, &entry);
> > +
> > +	if (status != HV_STATUS_SUCCESS) {
> > +		pr_err("%s: map hypercall failed, status %d\n", __func__, status);
> > +		return -1;
> 
> and here.
> 

-EINVAL would be more appropriate in both cases.

Wei.

> > +	}
> > +
> > +	/* Update the entry in mp_chip_data. It is used in other places. */
> > +	mp_data->entry = *(struct IO_APIC_route_entry *)&entry.ioapic_rte;
> > +
> > +	/* Sync polarity -- Hyper-V's returned polarity is always 0... */
> > +	mp_data->entry.polarity = ire.polarity;
> > +
> > +	eu.w1 = entry.ioapic_rte.low_uint32;
> > +	eu.w2 = entry.ioapic_rte.high_uint32;
> > +	ioapic_write_entry(ioapic, ioapic_pin, eu.entry);
> > +
> > +	return 0;
> > +}
> > +
> > +static void hv_ioapic_mask_irq(struct irq_data *data)
> > +{
> > +	mask_ioapic_irq(data);
> > +}
> > +
> > +static void hv_ioapic_unmask_irq(struct irq_data *data)
> > +{
> > +	unmask_ioapic_irq(data);
> > +}
> > +
> > +static int hv_ioapic_set_affinity(struct irq_data *data,
> > +			       const struct cpumask *mask, bool force)
> > +{
> > +	/*
> > +	 * We only update the affinity mask here. Programming the hardware is
> > +	 * done in irq_startup.
> > +	 */
> > +	return ioapic_set_affinity(data, mask, force);
> > +}
> > +
> > +void hv_ioapic_ack_level(struct irq_data *irq_data)
> > +{
> > +	/*
> > +	 * Per email exchange with Hyper-V team, all is needed is write to
> > +	 * LAPIC's EOI register. They don't support directed EOI to IO-APIC.
> > +	 * Hyper-V handles it for us.
> > +	 */
> > +	apic_ack_irq(irq_data);
> > +}
> > +
> > +struct irq_chip hv_ioapic_chip __read_mostly = {
> > +	.name			= "HV-IO-APIC",
> > +	.irq_startup		= hv_ioapic_startup_irq,
> > +	.irq_mask		= hv_ioapic_mask_irq,
> > +	.irq_unmask		= hv_ioapic_unmask_irq,
> > +	.irq_ack		= irq_chip_ack_parent,
> > +	.irq_eoi		= hv_ioapic_ack_level,
> > +	.irq_set_affinity	= hv_ioapic_set_affinity,
> > +	.irq_retrigger		= irq_chip_retrigger_hierarchy,
> > +	.irq_get_irqchip_state	= ioapic_irq_get_chip_state,
> > +	.flags			= IRQCHIP_SKIP_SET_WAKE,
> > +};
> > +
> > +
> > +int (*native_acpi_register_gsi)(struct device *dev, u32 gsi, int trigger, int polarity);
> > +void (*native_acpi_unregister_gsi)(u32 gsi);
> > +
> > +int hv_acpi_register_gsi(struct device *dev, u32 gsi, int trigger, int polarity)
> > +{
> > +	int irq = gsi;
> > +
> > +#ifdef CONFIG_X86_IO_APIC
> > +	irq = native_acpi_register_gsi(dev, gsi, trigger, polarity);
> > +	if (irq < 0) {
> > +		pr_err("native_acpi_register_gsi failed %d\n", irq);
> > +		return irq;
> > +	}
> > +
> > +	if (trigger) {
> > +		irq_set_status_flags(irq, IRQ_LEVEL);
> > +		irq_set_chip_and_handler_name(irq, &hv_ioapic_chip,
> > +			handle_fasteoi_irq, "ioapic-fasteoi");
> > +	} else {
> > +		irq_clear_status_flags(irq, IRQ_LEVEL);
> > +		irq_set_chip_and_handler_name(irq, &hv_ioapic_chip,
> > +			handle_edge_irq, "ioapic-edge");
> > +	}
> > +#endif
> > +	return irq;
> > +}
> > +
> > +void hv_acpi_unregister_gsi(u32 gsi)
> > +{
> > +#ifdef CONFIG_X86_IO_APIC
> > +	(void)hv_unmap_ioapic_interrupt(gsi);
> > +	native_acpi_unregister_gsi(gsi);
> > +#endif
> > +}
> 
> -- 
> Vitaly
>