于 2013/3/6 11:54, Michael Ellerman 写道:
> On Tue, Mar 05, 2013 at 03:19:57PM +0800, Mike Qiu wrote:
>> 于 2013/3/5 10:23, Michael Ellerman 写道:
>>> On Tue, Jan 15, 2013 at 03:38:55PM +0800, Mike Qiu wrote:
>>>> Adding a function irq_create_mapping_many() which can associate
>>>> multiple MSIs to a continous irq mapping.
>>>>
>>>> This is needed to enable multiple MSI support for pSeries.
>>>>
>>>> Signed-off-by: Mike Qiu <qiudayu@linux.vnet.ibm.com>
>>>> ---
>>>>   include/linux/irq.h       |    2 +
>>>>   include/linux/irqdomain.h |    3 ++
>>>>   kernel/irq/irqdomain.c    |   61 +++++++++++++++++++++++++++++++++++++++++++++
>>>>   3 files changed, 66 insertions(+), 0 deletions(-)
>>>>
>>>> diff --git a/include/linux/irq.h b/include/linux/irq.h
>>>> index 60ef45b..e00a7ec 100644
>>>> --- a/include/linux/irq.h
>>>> +++ b/include/linux/irq.h
>>>> @@ -592,6 +592,8 @@ int __irq_alloc_descs(int irq, unsigned int from, unsigned int cnt, int node,
>>>>   #define irq_alloc_desc_from(from, node)		\
>>>>   	irq_alloc_descs(-1, from, 1, node)
>>>> +#define irq_alloc_desc_n(nevc, node)		\
>>>> +	irq_alloc_descs(-1, 0, nevc, node)
>>> This has been superseeded by irq_alloc_descs_from(), which is the right
>>> way to do it.
>> Yes, but irq_alloc_descs_from() just for 1 irq
> No it's not, look again.
>
> #define irq_alloc_descs_from(from, cnt, node)   \
> 	irq_alloc_descs(-1, from, cnt, node)
Sorry, I see as irq_alloc_desc_from(from, node)
you are right
>
>
>>>> diff --git a/kernel/irq/irqdomain.c b/kernel/irq/irqdomain.c
>>>> index 96f3a1d..38648e6 100644
>>>> --- a/kernel/irq/irqdomain.c
>>>> +++ b/kernel/irq/irqdomain.c
>>>> @@ -636,6 +636,67 @@ int irq_create_strict_mappings(struct irq_domain *domain, unsigned int irq_base,
>>>>   }
>>>>   EXPORT_SYMBOL_GPL(irq_create_strict_mappings);
>>>> +/**
>>>> + * irq_create_mapping_many - Map a range of hw IRQs to a range of virtual IRQs
>>>> + * @domain: domain owning the interrupt range
>>>> + * @hwirq_base: beginning of continuous hardware IRQ range
>>>> + * @count: Number of interrupts to map
>>> For multiple-MSI the allocated interrupt numbers must be a power-of-2,
>>> and must be naturally aligned. I don't /think/ that's a requirement for
>>> the virtual numbers, but it's probably best that we do it anyway.
>>>
>>> So this API needs to specify that it will give you back a power-of-2
>>> block that is naturally aligned - otherwise you can't use it for MSI.
>> rtas_call will return the numbers of hardware interrupt, and it
>> should be power-of-2, as this I think do not need to specify
> You're confusing hardware interrupt numbers and virtual interrupt
> numbers. My comment is about irq_create_mapping_many(), which returns
> virtual interrupt numbers.
>
> As I said I don't think there is a requirement that the virtual
> interrupt numbers are also a power-of-2 naturally aligned block, but we
> should allocate them as one anyway, to avoid any issues in future.
But for virtual interrupt numbersit should be a power-of-2 naturally
aligned block, because it must be continuous, as the MSI-HOWTO.txt says:

     4.2.2 pci_enable_msi_block
     int pci_enable_msi_block(struct pci_dev *dev, int count)
     This variation on the above call allows a device driver to request
     multiple MSIs.  The MSI specification only allows interrupts to be
     allocated in powers of two, up to a maximum of 2^5 (32).
     If this function returns 0, it has succeeded in allocating at least
     as many interrupts as the driver requested
     (it may have allocated more in order to satisfy the power-of-two
     requirement). In this case, the function enables MSI on this device
     and updates dev->irq to be the lowest of the new interrupts
     assigned to it. The other interrupts assigned to the device are in
     the range dev->irq to dev->irq + count - 1.

See the last line, that means for the virtual interrupts must be a
continuous block.
> And so this API, which returns virtual interrupt numbers, must satisfy
> that specification.
>
>>>> +	/* Look for default domain if nececssary */
>>>> +	if (!domain)
>>>> +		domain = irq_default_domain;
>>>> +	if (!domain) {
>>>> +		pr_warn("irq_create_mapping called for NULL domain, hwirq=%lx\n"
>>>> +			, hwirq_base);
>>>> +		WARN_ON(1);
>>>> +		return 0;
>>>> +	}
>>>> +	pr_debug("-> using domain @%p\n", domain);
>>>> +
>>>> +	/* For IRQ_DOMAIN_MAP_LEGACY, get the first virtual interrupt number */
>>>> +	if (domain->revmap_type == IRQ_DOMAIN_MAP_LEGACY)
>>>> +		return irq_domain_legacy_revmap(domain, hwirq_base);
>>> The above doesn't work.
>> Why it doesn't work ?
> Because irq_domain_legacy_revmap() only allocates a single interrupt
> number.
OK, your right.
>>>> +	/* Check if mapping already exists */
>>>> +	for (i = 0; i < count; i++) {
>>>> +		virq = irq_find_mapping(domain, hwirq_base+i);
>>>> +		if (virq) {
>>>> +			pr_debug("existing mapping on virq %d,"
>>>> +					" now dispose it first\n", virq);
>>>> +			irq_dispose_mapping(virq);
>>> You might have just disposed of someone elses mapping, we shouldn't do
>>> that. It should be an error to the caller.
>> It's a good question. If the interrupt used for someone elses, why I
>> can apply it from the system?
> I agree, that would be a bug. But disposing of someone elses mapping is
> not OK.
>
>> So it may someone else forget to dispose mapping, and it never be
>> used for others as I have got the interrupt I think.
> Perhaps, but that is a bug that needs to be fixed in the code that
> forgets to dispose of the mapping.
>
> cheers
>