Re: [Qemu-devel] [PATCH v19 3/7] xbitmap: add more operations

[Matthew Wilcox]

On Sun, Dec 17, 2017 at 01:47:21PM +0000, Wang, Wei W wrote:
> On Saturday, December 16, 2017 3:22 AM, Matthew Wilcox wrote:
> > On Fri, Dec 15, 2017 at 10:49:15AM -0800, Matthew Wilcox wrote:
> > > Here's the API I'm looking at right now.  The user need take no lock;
> > > the locking (spinlock) is handled internally to the implementation.
> 
> Another place I saw your comment " The xb_ API requires you to handle your 
> own locking" which seems conflict with the above "the user need take no lock".
> Doesn't the caller need a lock to avoid concurrent accesses to the ida bitmap?

Yes, the xb_ implementation requires you to handle your own locking.
The xbit_ API that I'm proposing will take care of the locking for you.
There's also no preallocation in the API.

> We'll change it to "bool xb_find_set(.., unsigned long *result)", returning 
> false indicates no "1" bit is found.

I put a replacement proposal in the next paragraph:
bool xbit_find_set(struct xbitmap *, unsigned long *start, unsigned long max);

Maybe 'start' is the wrong name for that parameter.  Let's call it 'bit'.
It's both "where to start" and "first bit found".

> >  - xbit_clear() can't return an error.  Neither can xbit_zero().
> 
> I found the current xbit_clear implementation only returns 0, and there isn't 
> an error to be returned from this function. In this case, is it better to 
> make the function "void"?

Yes, I think so.

My only qualm is that I've been considering optimising the memory
consumption when an entire 1024-bit chunk is full; instead of keeping a
pointer to a 128-byte entry full of ones, store a special value in the
radix tree which means "every bit is set".

The downside is that we then have to pass GFP flags to xbit_clear() and
xbit_zero(), and they can fail.  It's not clear to me whether that's a
good tradeoff.

> Are you suggesting to rename the current xb_ APIs to the above xbit_ names 
> (with parameter changes)? 
> 
> Why would we need xbit_alloc, which looks like ida_get_new, I think set/clear 
> should be adequate to the current usages.

I'm intending on replacing the xb_ and ida_ implementations with this one.
It removes the preload API which makes it easier to use, and it handles
the locking for you.

But I need to get the XArray (which replaces the radix tree) finished first.