Re: [Xen-devel] [PATCH] drivers/block/xen-blkback/common.h: use DIV_ROUND_UP instead of reimplementing its function

[Julien Grall <julien.grall@arm.com>]

Hi,

On 09/12/2018 10:16 AM, Roger Pau Monné wrote:
> On Wed, Sep 12, 2018 at 11:13:50AM +0200, Roger Pau Monné wrote:
>> Adding Julien how did the work to support XEN_PAGE_SIZE != PAGE_SIZE.
>>
>> On Wed, Sep 12, 2018 at 02:14:26AM -0600, Jan Beulich wrote:
>>>>>> On 12.09.18 at 07:45, <zhongjiang@huawei.com> wrote:
>>>> --- a/drivers/block/xen-blkback/common.h
>>>> +++ b/drivers/block/xen-blkback/common.h
>>>> @@ -65,7 +65,7 @@
>>>>   	(XEN_PAGES_PER_INDIRECT_FRAME / XEN_PAGES_PER_SEGMENT)
>>>>   
>>>>   #define MAX_INDIRECT_PAGES \
>>>> -	((MAX_INDIRECT_SEGMENTS + SEGS_PER_INDIRECT_FRAME - 1)/SEGS_PER_INDIRECT_FRAME)
>>>> +		DIV_ROUND_UP(MAX_INDIRECT_SEGMENTS, SEGS_PER_INDIRECT_FRAME)
>>>>   #define INDIRECT_PAGES(_segs) DIV_ROUND_UP(_segs, XEN_PAGES_PER_INDIRECT_FRAME)
>>>
>>> My first reaction was to suggest
>>>
>>> #define MAX_INDIRECT_PAGES INDIRECT_PAGES(MAX_INDIRECT_SEGMENTS)
>>>
>>> but that wouldn't match what's there currently (note the two different
>>> divisors). I can't really decide whether that's just unfortunate naming
>>> of the two macros, or an actual bug.
>>
>> I think there's indeed a bug here.
>>
>> AFAICT, MAX_INDIRECT_PAGES should use XEN_PAGES_PER_INDIRECT_FRAME and
>> then it could be changed as Jan suggested.

The problem is SEGS_PER_INDIRECT_FRAME has been miscalculated. So I 
think it would be fine to use XEN_PAGES_PER_INDIRECT_FRAME in 
MAX_INDIRECT_PAGES.

However the naming for XEN_PAGES_PER_INDIRECT_FRAME is misnamed. We 
return number of a for segments per indirect frame. So I would rename to 
SEGS_PER_INDIRECT_FRAME.

>>
>> Current MAX_INDIRECT_PAGES is misnamed and should instead be
>> MAX_INDIRECT_SEGS (which on x86 is exactly the same because PAGE_SIZE
>> == XEN_PAGE_SIZE).

Looking at the usage:

j = min(MAX_INDIRECT_PAGES, INDIRECT_PAGES(nr_segments))

Where j is used as the number of grant ref. So I don't think the 
variable is misnamed here.

Cheers,

-- 
Julien Grall