[libvirt] [PATCH v5 06/13] conf: Introduce address caching for PCI extensions

[Yi Min Zhao]

在 2018/9/12 下午6:37, Andrea Bolognani 写道:
> On Wed, 2018-09-12 at 16:34 +0800, Yi Min Zhao wrote:
>> 在 2018/9/12 下午3:35, Yi Min Zhao 写道:
>>>> This makes sense and seems to work just fine; however, you are
>>>> allocating and releasing a bunch of small integers, which seems
>>>> a bit wasteful.
>>>>
>>>> vircgroup is AFAICT avoiding all that extra memory management by
>>>> stuffing the values straight into the pointers themselves, which
>>>> you should also be able to do since the biggest legal ID is a
>>>> 32-bit integer.
>>>>
>>>> That said, I haven't been able to get that to actually work, at
>>>> least with a quick attempt :( Would you mind exploring that route
>>>> and figuring out whether it's feasible at all?
>>> I'm testing this. Actually I wanted to do so like vircgroup. I
>>> remembered there's
>>> error due to the previous code logic. I will reply to you later.
>> I remebered the reason and test again. FID might be 0. It is treated as
>> an error
>> if we save 0 in void* pointer.
> Right.
>
> Too bad fid can go all the way to UINT32_MAX, otherwise we could
> have just stored them in the pointer after offsetting them by one
> and thus worked around the issue...
Yes. Just one value makes all things complex.
>
> I guess forbidding users from using UINT32_MAX as a fid is not an
> option, right?
>
Actually as my understanding, it's just a value to identify the pci 
function.
IMO, it's not a big deal to decrease usable FID values. After all, UID set
is smaller than FID set. The maximum number of pci devices is limited
by UID. Anyway, I have to discuss this with my colleagues internally.
I will tell you our discussion result first time.

-- 
Yi Min