Re: [Xen-devel] xen/arm: gicv3: Computation of GICD_TYPER.IDbits

[Julien Grall <julien.grall@xxxxxxxxxx>]

On 18/09/15 10:48, Vijay Kilari wrote:
> On Fri, Sep 18, 2015 at 2:03 PM, Ian Campbell <ian.campbell@xxxxxxxxxx> wrote:
>> On Fri, 2015-09-18 at 12:56 +0530, Vijay Kilari wrote:
>>> Hi Julien,
>>>
>>>  static int vgic_v3_distr_mmio_read(struct vcpu *v, mmio_info_t *info)
>>>  {
>>>      ...
>>>            case GICD_TYPER:
>>>            {
>>>                /*
>>>                 * Number of interrupt identifier bits supported by the
>>> GIC
>>>                 * Stream Protocol Interface
>>>                 */
>>>                unsigned int irq_bits =
>>> get_count_order(vgic_num_irqs(v->domain));
>>> }
>>>
>>> I think we should use ilog2() instead of get_count_order()?
>>
>> Please explain why.
> 
> For a given value, get_count_order returns next highest 2^n bit that can 
> contain
> this number, where as ilog2 returns number max highest bit set to
> contain the value.
> (Here is the output of get_count_order() and ilog2() for various sizes )
> 
> 128K get_count_order 17 log2 17
> 64K get_count_order   16 log2 16
> 48K get_count_order   16 log2 15

2^16 = 64K
2^15 = 32K

So here the number of bits returned wouldn't be enough to fit the value.
We would expose the wrong number of IRQ to the guest.

> 32K get_count_order   15 log2 15
> 24K get_count_order   15 log2 14
> 16K get_count_order   14 log2 14
> 8K get_count_order     13 log2 13
> 
> Here if irq_bits should specify number of bits required to specify the number.
> So log2 is more appropriate. Even Linux ITS driver uses ilog2 to count number
> of bits required.

ilog2 is used in the ITS code because they round up the number of ITEs
to a power of 2.

ilog2 will always return the highest bit set in the number. So if it's
not a power of two it will return one bit less than it's required to
contain the value.

get_count_order will take care of it by adding 1 if the number is not a
power of 2.

Regards,

-- 
Julien Grall

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