Re: [RFC PATCH 04/10] pipe: Use head and tail pointers for the ring, not cursor and length [ver #2]

[Rasmus Villemoes <linux@rasmusvillemoes.dk>]

On 30/10/2019 17.19, Ilya Dryomov wrote:
> On Thu, Oct 24, 2019 at 11:49 AM David Howells <dhowells@redhat.com> wrote:
>>  /*
>> - * We use a start+len construction, which provides full use of the
>> - * allocated memory.
>> - * -- Florian Coosmann (FGC)
>> - *
>> + * We use head and tail indices that aren't masked off, except at the point of
>> + * dereference, but rather they're allowed to wrap naturally.  This means there
>> + * isn't a dead spot in the buffer, provided the ring size < INT_MAX.
>> + * -- David Howells 2019-09-23.
> 
> Hi David,
> 
> Is "ring size < INT_MAX" constraint correct?

No. As long as one always uses a[idx % size] to access the array, the
only requirement is that size is representable in an unsigned int. Then
because one also wants to do the % using simple bitmasking, that further
restricts one to sizes that are a power of 2, so the end result is that
the max size is 2^31 (aka INT_MAX+1).

> I've never had to implement this free running indices scheme, but
> the way I've always visualized it is that the top bit of the index is
> used as a lap (as in a race) indicator, leaving 31 bits to work with
> (in case of unsigned ints).  Should that be
> 
>   ring size <= 2^31
> 
> or more precisely
> 
>   ring size is a power of two <= 2^31

Exactly. But it's kind of moot since the ring size would never be
allowed to grow anywhere near that.

Rasmus