[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <150470157212.5486.15223573670024117741@mail.alporthouse.com>
Date: Wed, 06 Sep 2017 13:39:32 +0100
From: Chris Wilson <chris@...is-wilson.co.uk>
To: Tvrtko Ursulin <tvrtko.ursulin@...ux.intel.com>,
"Tvrtko Ursulin" <tursulin@...ulin.net>,
Intel-gfx@...ts.freedesktop.org
Cc: linux-kernel@...r.kernel.org
Subject: Re: [Intel-gfx] [PATCH v2 5/5] tools/testing/scatterlist: Test new
__sg_alloc_table_from_pages
Quoting Tvrtko Ursulin (2017-09-06 13:10:57)
>
> On 06/09/2017 11:48, Chris Wilson wrote:
> > All ascending. Interesting challenge for 3,2,1,0; it can be coalesced,
> > we just don't. I wonder if we are missing some like that. But for the
>
> Hm, how do you think descending pages could be coalesced? By
> re-arranging the pages? But that would break everything, do I don't get it.
Wishful thinking; I wasn't considering the order inside the object, just
their physical addresses.
>
> > moment, 0, 2, 1, would be a good addition to the above set.
> >
> > Is there any value in checking overflows in this interface?
>
> Overflows as in size > num_pages * PAGE_SIZE as passed in to
> __sg_alloc_table_from_pages ? It is not checked in the implementation at
> the moment and it looks it is harmless.
Just thinking aloud if there was a way to get a mult/add overflow. That
we do page size coalescing, the only avenue is from a buggy max_seg.
Going back to the makefile, perhaps add the magic for ubsan as well?
-fsanitize=address -fsanitize=undefined
-Chris
Powered by blists - more mailing lists