Re: [PATCH 1/1] x86/apic: Introduce paravirq irq_domain

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 12.08.2016 14:43, Paolo Bonzini wrote:
> On 12/08/2016 12:56, Alexander Popov wrote:
>> Maybe the name "paravirq" is not very good, I'll try to describe the idea.
>>
>> There is some kernel module for special interactions between guest VMs.
>> Currently it has to register a MSI-capable PCI device to handle interrupts
>> injected by the hypervisor. And the bare-metal hypervisor has to emulate
>> such a device for guest VMs.
>>
>> So I've implemented paravirq irq_domain to avoid this redundant emulation.
>> With it we can just call:
>>  - paravirq_alloc_irq() to allocate a LAPIC irq;
>>  - request_irq() for it;
>>  - irqd_cfg(irq_get_irq_data()) to get the corresponding interrupt vector
>> and inform the hypervisor about it.
>> Now we happily handle the irq from the hypervisor when it injects this vector.
>>
>> The irq_mask/irq_unmask parameters of paravirq_init_chip() are the pointers
>> to the functions from the interaction module which ask the hypervisor to
>> start/stop injecting interrupts to the guest VM.
>>
>> Paravirq irq_domain allows to avoid the PCI device emulation in the hypervisor
>> and provides the ability to run slimmer Linux guests without precompiled
>> PCI and MSI support.
>>
>> Did I manage to answer your questions?
> 
> It's a bit clearer.  My doubt is that the caller of paravirq_init_chip
> has to provide irq_mask and irq_unmask, but it doesn't know who will
> call paravirq_alloc_irq.  So there are two cases:
> 
> 1) there is only one device, and then your solution doesn't scale well
> to multiple devices
> 
> 2) there is some kind of commonality between all devices using
> paravirq_alloc_irq, and then it should be abstracted in a bus.
> 
> The latter would be similar to what Xen and Hyper-V do, for example.
> Using PCI is more similar to the KVM approach.

Excuse me, I don't see the problem.

The caller of paravirq_init_chip() provides irq_mask/irq_unmask
function pointers only once, and paravirq_init_chip() saves them in
.irq_mask/.irq_unmask fields of struct irq_chip paravirq_chip.

When later, for example, disable_irq() is called for one of several irqs
allocated in paravirq irq_domain, paravirq_chip->irq_mask() is called
with struct irq_desc *desc argument corresponding to that particular irq.

I.e. our irq_mask()/irq_unmask() callbacks get irq_desc of the interrupt
which should be masked/unmasked and can ask the hypervisor to stop/start
injecting the vector of that particular interrupt.


>>>> Signed-off-by: Alexander Popov <alex.popov@xxxxxxxxx>
>>>> ---
>>>>  arch/x86/Kconfig                 |   8 +++
>>>>  arch/x86/include/asm/irqdomain.h |   6 ++
>>>>  arch/x86/include/asm/paravirq.h  |   9 +++
>>>>  arch/x86/kernel/apic/Makefile    |   2 +
>>>>  arch/x86/kernel/apic/paravirq.c  | 128
>>>>  +++++++++++++++++++++++++++++++++++++++
>>>>  arch/x86/kernel/apic/vector.c    |   1 +
>>>>  6 files changed, 154 insertions(+)
>>>>  create mode 100644 arch/x86/include/asm/paravirq.h
>>>>  create mode 100644 arch/x86/kernel/apic/paravirq.c
>>>>
>>>> diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
>>>> index 5c6e747..209bd88 100644
>>>> --- a/arch/x86/Kconfig
>>>> +++ b/arch/x86/Kconfig
>>>> @@ -760,6 +760,14 @@ config PARAVIRT_TIME_ACCOUNTING
>>>>  
>>>>  	  If in doubt, say N here.
>>>>  
>>>> +config X86_PARAVIRQ
>>>> +	bool "Enable paravirq irq_domain"
>>>> +	depends on PARAVIRT && X86_LOCAL_APIC
>>>> +	default n
>>>> +	---help---
>>>> +	  This option enables paravirq irq_domain for interrupts injected
>>>> +	  by the hypervisor using Intel VT-x technology.
>>>> +
>>>>  config PARAVIRT_CLOCK
>>>>  	bool
>>>>  
>>>> diff --git a/arch/x86/include/asm/irqdomain.h
>>>> b/arch/x86/include/asm/irqdomain.h
>>>> index d26075b..e3192f6 100644
>>>> --- a/arch/x86/include/asm/irqdomain.h
>>>> +++ b/arch/x86/include/asm/irqdomain.h
>>>> @@ -60,4 +60,10 @@ extern void arch_init_htirq_domain(struct irq_domain
>>>> *domain);
>>>>  static inline void arch_init_htirq_domain(struct irq_domain *domain) { }
>>>>  #endif
>>>>  
>>>> +#ifdef CONFIG_X86_PARAVIRQ
>>>> +extern void arch_init_paravirq_domain(struct irq_domain *domain);
>>>> +#else
>>>> +static inline void arch_init_paravirq_domain(struct irq_domain *domain) { }
>>>> +#endif
>>>> +
>>>>  #endif
>>>> diff --git a/arch/x86/include/asm/paravirq.h
>>>> b/arch/x86/include/asm/paravirq.h
>>>> new file mode 100644
>>>> index 0000000..a137de2
>>>> --- /dev/null
>>>> +++ b/arch/x86/include/asm/paravirq.h
>>>> @@ -0,0 +1,9 @@
>>>> +#ifndef _ASM_X86_PARAVIRQ_H
>>>> +#define _ASM_X86_PARAVIRQ_H
>>>> +
>>>> +int paravirq_init_chip(void (*irq_mask)(struct irq_data *data),
>>>> +				void (*irq_unmask)(struct irq_data *data));
>>>> +int paravirq_alloc_irq(void);
>>>> +void paravirq_free_irq(unsigned int irq);
>>>> +
>>>> +#endif /* _ASM_X86_PARAVIRQ_H */
>>>> diff --git a/arch/x86/kernel/apic/Makefile b/arch/x86/kernel/apic/Makefile
>>>> index 8e63ebd..84f9ce0 100644
>>>> --- a/arch/x86/kernel/apic/Makefile
>>>> +++ b/arch/x86/kernel/apic/Makefile
>>>> @@ -28,3 +28,5 @@ obj-$(CONFIG_X86_BIGSMP)	+= bigsmp_32.o
>>>>  
>>>>  # For 32bit, probe_32 need to be listed last
>>>>  obj-$(CONFIG_X86_LOCAL_APIC)	+= probe_$(BITS).o
>>>> +
>>>> +obj-$(CONFIG_X86_PARAVIRQ)	+= paravirq.o
>>>> diff --git a/arch/x86/kernel/apic/paravirq.c
>>>> b/arch/x86/kernel/apic/paravirq.c
>>>> new file mode 100644
>>>> index 0000000..430b819
>>>> --- /dev/null
>>>> +++ b/arch/x86/kernel/apic/paravirq.c
>>>> @@ -0,0 +1,128 @@
>>>> +/*
>>>> + * An irq_domain for interrupts injected by the hypervisor using
>>>> + * Intel VT-x technology.
>>>> + *
>>>> + * Copyright (C) 2016 Alexander Popov <alex.popov@xxxxxxxxx>.
>>>> + *
>>>> + * This file is released under the GPLv2.
>>>> + */
>>>> +
>>>> +#include <linux/init.h>
>>>> +#include <linux/irq.h>
>>>> +#include <asm/irqdomain.h>
>>>> +#include <asm/paravirq.h>
>>>> +
>>>> +static struct irq_domain *paravirq_domain;
>>>> +
>>>> +static struct irq_chip paravirq_chip = {
>>>> +	.name			= "PARAVIRQ",
>>>> +	.irq_ack		= irq_chip_ack_parent,
>>>> +};
>>>> +
>>>> +static int paravirq_domain_alloc(struct irq_domain *domain,
>>>> +			unsigned int virq, unsigned int nr_irqs, void *arg)
>>>> +{
>>>> +	int ret = 0;
>>>> +
>>>> +	BUG_ON(domain != paravirq_domain);
>>>> +
>>>> +	if (nr_irqs != 1)
>>>> +		return -EINVAL;
>>>> +
>>>> +	ret = irq_domain_set_hwirq_and_chip(paravirq_domain,
>>>> +					virq, virq, &paravirq_chip, NULL);
>>>> +	if (ret) {
>>>> +		pr_warn("setting chip, hwirq for irq %u failed\n", virq);
>>>> +		return ret;
>>>> +	}
>>>> +
>>>> +	__irq_set_handler(virq, handle_edge_irq, 0, "edge");
>>>> +
>>>> +	return 0;
>>>> +}
>>>> +
>>>> +static void paravirq_domain_free(struct irq_domain *domain,
>>>> +					unsigned int virq, unsigned int nr_irqs)
>>>> +{
>>>> +	struct irq_data *irq_data;
>>>> +
>>>> +	BUG_ON(domain != paravirq_domain);
>>>> +	BUG_ON(nr_irqs != 1);
>>>> +
>>>> +	irq_data = irq_domain_get_irq_data(paravirq_domain, virq);
>>>> +	if (irq_data)
>>>> +		irq_domain_reset_irq_data(irq_data);
>>>> +	else
>>>> +		pr_warn("irq %u is not in paravirq irq_domain\n", virq);
>>>> +}
>>>> +
>>>> +static const struct irq_domain_ops paravirq_domain_ops = {
>>>> +	.alloc	= paravirq_domain_alloc,
>>>> +	.free	= paravirq_domain_free,
>>>> +};
>>>> +
>>>> +int paravirq_alloc_irq(void)
>>>> +{
>>>> +	struct irq_alloc_info info;
>>>> +
>>>> +	if (!paravirq_domain)
>>>> +		return -ENODEV;
>>>> +
>>>> +	if (!paravirq_chip.irq_mask || !paravirq_chip.irq_unmask)
>>>> +		return -EINVAL;
>>>> +
>>>> +	init_irq_alloc_info(&info, NULL);
>>>> +
>>>> +	return irq_domain_alloc_irqs(paravirq_domain, 1, NUMA_NO_NODE, &info);
>>>> +}
>>>> +EXPORT_SYMBOL(paravirq_alloc_irq);
>>>> +
>>>> +void paravirq_free_irq(unsigned int virq)
>>>> +{
>>>> +	struct irq_data *irq_data;
>>>> +
>>>> +	if (!paravirq_domain) {
>>>> +		pr_warn("paravirq irq_domain is not initialized\n");
>>>> +		return;
>>>> +	}
>>>> +
>>>> +	irq_data = irq_domain_get_irq_data(paravirq_domain, virq);
>>>> +	if (irq_data)
>>>> +		irq_domain_free_irqs(virq, 1);
>>>> +	else
>>>> +		pr_warn("irq %u is not in paravirq irq_domain\n", virq);
>>>> +}
>>>> +EXPORT_SYMBOL(paravirq_free_irq);
>>>> +
>>>> +int paravirq_init_chip(void (*irq_mask)(struct irq_data *data),
>>>> +				void (*irq_unmask)(struct irq_data *data))
>>>> +{
>>>> +	if (!paravirq_domain)
>>>> +		return -ENODEV;
>>>> +
>>>> +	if (paravirq_chip.irq_mask || paravirq_chip.irq_unmask)
>>>> +		return -EEXIST;
>>>> +
>>>> +	if (!irq_mask || !irq_unmask)
>>>> +		return -EINVAL;
>>>> +
>>>> +	paravirq_chip.irq_mask = irq_mask;
>>>> +	paravirq_chip.irq_unmask = irq_unmask;
>>>> +
>>>> +	return 0;
>>>> +}
>>>> +EXPORT_SYMBOL(paravirq_init_chip);
>>>> +
>>>> +void arch_init_paravirq_domain(struct irq_domain *parent)
>>>> +{
>>>> +	paravirq_domain = irq_domain_add_tree(NULL, &paravirq_domain_ops, NULL);
>>>> +	if (!paravirq_domain) {
>>>> +		pr_warn("failed to initialize paravirq irq_domain\n");
>>>> +		return;
>>>> +	}
>>>> +
>>>> +	paravirq_domain->name = paravirq_chip.name;
>>>> +	paravirq_domain->parent = parent;
>>>> +	paravirq_domain->flags |= IRQ_DOMAIN_FLAG_AUTO_RECURSIVE;
>>>> +}
>>>> +
>>>> diff --git a/arch/x86/kernel/apic/vector.c b/arch/x86/kernel/apic/vector.c
>>>> index 6066d94..878b440 100644
>>>> --- a/arch/x86/kernel/apic/vector.c
>>>> +++ b/arch/x86/kernel/apic/vector.c
>>>> @@ -438,6 +438,7 @@ int __init arch_early_irq_init(void)
>>>>  
>>>>  	arch_init_msi_domain(x86_vector_domain);
>>>>  	arch_init_htirq_domain(x86_vector_domain);
>>>> +	arch_init_paravirq_domain(x86_vector_domain);
>>>>  
>>>>  	BUG_ON(!alloc_cpumask_var(&vector_cpumask, GFP_KERNEL));
>>>>  	BUG_ON(!alloc_cpumask_var(&vector_searchmask, GFP_KERNEL));
>>>> --
>>>> 2.5.5
>>>>
>>>>
>>

--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html



[Index of Archives]     [KVM ARM]     [KVM ia64]     [KVM ppc]     [Virtualization Tools]     [Spice Development]     [Libvirt]     [Libvirt Users]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite Questions]     [Linux Kernel]     [Linux SCSI]     [XFree86]
  Powered by Linux