[lkml]   [2017]   [Mar]   [20]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRE: [PATCH 07/17] net: convert sock.sk_refcnt from atomic_t to refcount_t
From: Peter Zijlstra
> Sent: 20 March 2017 14:28
> On Mon, Mar 20, 2017 at 02:10:24PM +0000, David Laight wrote:
> > On x86 the cpu flags from the 'lock inc/dec' could be used to reasonably
> > cheaply detect errors - provided you actually generate a forwards branch.
> Note that currently there is no arch specific implementation. We could
> of course cure this.
> But note that the thing you propose; using the overflow flag, can only
> reasonably be done on PREEMPT=n kernels, otherwise we have an incredible
> number of contexts that can nest.
> Sure; getting all starts aligned to double overflow is incredibly rare,
> but I don't want to be the one to have to debug that.

One overflow would set the overflow flag, you don't need both to fail.

In any case you can use the sign flag.
Say valid count values are -64k to -256 and 0 to MAXINT.
The count will normally be +ve unless the 'main free path'
has released the 64k references it holds.
If the sign bit is set after inc/dec the value is checked;
might valid, an error, or require the item be freed.

Ok assuming the items have reasonable lifetimes and have a nominal
'delete' function.


 \ /
  Last update: 2017-03-20 16:06    [W:0.131 / U:6.748 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site