On Fri, Jul 13, 2012 at 2:38 AM, Octavian Purdila
<octavian.purd...@intel.com> wrote:
> On Thu, Jul 12, 2012 at 10:02 PM, Bjorn Helgaas <bhelg...@google.com> wrote:
>
>> Why don't you fix this right where the problem occurs, in
>> __reserve_region_with_split(), with something like this:
>>
>> if (end > conflict->start && conflict->start > start)
>> __reserve_region_with_split(root, start,
>> conflict->start-1, name);
>> if (start < conflict->end && conflict->end < end && )
>> __reserve_region_with_split(root, conflict->end+1, end,
>> name);
>>
>
> I am not sure this will behave properly in the overlapping cases.
> Consider start=10, end=150, root->start=100, root->end=200.
>
> In this case only the first condition above will be true (150 > 100 &&
> 100 > 10) and we will try to reserve [10, 99] - which will return
> nicely. But we will not reserve [100, 150].
OK, I see your point -- I was unclear on the semantics of
reserve_region_with_split(). Apparently it is supposed to reserve
both pieces, even if it crosses the edge of an existing region.
> But more important is the fact that by fixing the issue here we won't
> be able to log the error and give a chance to upper layer fixing the
> problem.
>
>>> + if (start > root->end || end < root->start) {
>>> + abort = 1;
>>> + pr_err("unable to fix request, aborting it\n");
>>
>> This message doesn't contain any useful information (range/root/etc).
>>
>
> Good point, I will remove it.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
