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

Re: [libvirt] [PATCH] Add VIR_DIV_UP to divide memory or storage request sizes with round up



2011/1/28 Eric Blake <eblake redhat com>:
> [replying to myself - never a good sign on Friday afternoon]
>
> On 01/28/2011 02:59 PM, Eric Blake wrote:
>>> +/* divide value by size, rounding up */
>>> +# define VIR_DIV_UP(value, size) (((value) + (size) - 1) / (size))
>>
>> Hmm - now that we are codifying it, can we make it less prone to
>> overflow issues?
>
>> ACK to everything else, but let's make sure we agree on the final
>> version of VIR_DIV_UP being committed.
>
> One possible issue is that pre-patch [if value is int and size is a
> power of 2], then INT_MAX / size * size is nicely truncated to INT_MAX &
> ~size, while post-patch, VIR_DIV_UP(INT_MAX, size) * size overflows.
> That is, while we can tweak your patch to avoid overflow on the division
> side, it doesn't affect overflow on the re-multiplication side, so
> avoiding overflow is a bigger audit anyways.
>
> That, and you've refactored it, so if we change our minds, it's now a
> single line to change (rather than every caller of VIR_DIV_UP being
> open-coded).
>
> So let's push it as-is (but with an additional tweak to cover the new
> divisions in 'virsh freecell --all' that were added in the meantime).
>

virsh freecell --all reports available memory. It doesn't deal with
memory requests. The point of rounding up requests is to ensure that
you actually get at least what you requested, maybe a bit more, but
not less.

When reporting available memory truncating the value is better than
rounding up, because rounding up might result in reporting more free
memory than there actually is.

As discussed on IRC I pushed the patch as-is now.

Matthias


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