Re: [Xen-devel] [PATCH v6 09/36] ARM: GICv3 ITS: introduce device mapping

[Andre Przywara <andre.przywara@xxxxxxx>]

Hi,

On 07/04/17 19:21, Stefano Stabellini wrote:
>> +    /*
>> +     * Allocate the pending_irqs for each virtual LPI. They will be put
>> +     * into the domain's radix tree upon the guest's MAPTI command.
>> +     * Pre-allocating memory for each *possible* LPI would be using way
>> +     * too much memory (they can be sparsely used by the guest), also
>> +     * allocating them on demand requires memory allocation in the interrupt
>> +     * injection code path, which is not really desired.
>> +     * So we compromise here by pre-allocating memory for each *mapped* LPI.
>> +     * See the mailing list discussion for some background:
>> +     * https://lists.xen.org/archives/html/xen-devel/2017-03/msg03645.html
> I don't mean to be picky, but I don't think this comment is entirely
> accurate: a "mapped LPI" should be an LPI for which MAPTI has been
> called, but actually here we are allocating pending_irq struct for every
> possible event specified by MAPD. Am I right?

Totally, and especially with the GIC and ITS it's important the get the
wording right.

> I would rephrase the last sentence to:
> 
>   So we compromise here by pre-allocating memory for each possible event
>   up to the max specified by MAPD.

Took that.

> With this comment fixed, you can add my reviewed-by.

Thanks!

Andre.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel