Message-ID: <20180119221243.GL13338@ZenIV.linux.org.uk>
Date:   Fri, 19 Jan 2018 22:12:43 +0000
From:   Al Viro <viro@...IV.linux.org.uk>
To:     Linus Torvalds <torvalds@...ux-foundation.org>
Cc:     Matthew Wilcox <willy@...radead.org>,
        "Kirill A. Shutemov" <kirill@...temov.name>,
        Peter Zijlstra <peterz@...radead.org>,
        Andrea Arcangeli <aarcange@...hat.com>,
        Dave Hansen <dave.hansen@...ux.intel.com>,
        Tetsuo Handa <penguin-kernel@...ove.sakura.ne.jp>,
        "Kirill A. Shutemov" <kirill.shutemov@...ux.intel.com>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Johannes Weiner <hannes@...xchg.org>,
        Joonsoo Kim <iamjoonsoo.kim@....com>,
        Mel Gorman <mgorman@...hsingularity.net>,
        Tony Luck <tony.luck@...el.com>,
        Vlastimil Babka <vbabka@...e.cz>,
        Michal Hocko <mhocko@...nel.org>,
        "hillf.zj" <hillf.zj@...baba-inc.com>,
        Hugh Dickins <hughd@...gle.com>,
        Oleg Nesterov <oleg@...hat.com>,
        Rik van Riel <riel@...hat.com>,
        Srikar Dronamraju <srikar@...ux.vnet.ibm.com>,
        Vladimir Davydov <vdavydov.dev@...il.com>,
        Ingo Molnar <mingo@...nel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        linux-mm <linux-mm@...ck.org>,
        the arch/x86 maintainers <x86@...nel.org>
Subject: Re: [mm 4.15-rc8] Random oopses under memory pressure.

On Fri, Jan 19, 2018 at 10:42:18AM -0800, Linus Torvalds wrote:
> On Fri, Jan 19, 2018 at 4:55 AM, Matthew Wilcox <willy@...radead.org> wrote:
> >
> > So really we should be casting 'b' and 'a' to uintptr_t to be fully
> > compliant with the spec.
> 
> That's an unnecessary technicality.
> 
> Any compiler that doesn't get pointer inequality testing right is not
> worth even worrying about. We wouldn't want to use such a compiler,
> because it's intentionally generating garbage just to f*ck with us.
> Why would you go along with that?
> 
> So the only real issue is that pointer subtraction case.
> 
> I actually asked (long long ago) for an optinal compiler warning for
> "pointer subtraction with non-power-of-2 sizes". Not because of it
> being undefined, but simply because it's expensive. The
> divide->multiply thing doesn't always work, and a real divide is
> really quite expensive on many architectures.
> 
> We *should* be careful about it. I guess sparse could be made to warn,
> but I'm afraid that we have so many of these things that a warning
> isn't reasonable.

You mean like -Wptr-subtraction-blows?

FWIW, allmodconfig on amd64 with C=2 CF=-Wptr-subtraction-blows is not too large
The tail (alphabetically sorted) is
lib/dynamic_debug.c:1013:9: warning: potentially expensive pointer subtraction
lib/extable.c:70:28: warning: potentially expensive pointer subtraction
mm/memory_hotplug.c:1530:13: warning: potentially expensive pointer subtraction
mm/memory_hotplug.c:734:13: warning: potentially expensive pointer subtraction
mm/memory_hotplug.c:831:41: warning: potentially expensive pointer subtraction
mm/page_owner.c:607:38: warning: potentially expensive pointer subtraction
mm/vmstat.c:1334:38: warning: potentially expensive pointer subtraction
net/core/net-sysfs.c:1040:19: warning: potentially expensive pointer subtraction
net/ipv4/ipmr.c:3026:32: warning: potentially expensive pointer subtraction
net/ipv6/ip6mr.c:458:32: warning: potentially expensive pointer subtraction
net/mac80211/tx.c:1307:41: warning: potentially expensive pointer subtraction
net/mac80211/tx.c:1351:44: warning: potentially expensive pointer subtraction
net/rds/ib_recv.c:345:49: warning: potentially expensive pointer subtraction
net/rds/ib_recv.c:861:38: warning: potentially expensive pointer subtraction
net/sched/cls_tcindex.c:622:43: warning: potentially expensive pointer subtraction
net/sched/sch_cbs.c:302:35: warning: potentially expensive pointer subtraction
net/sched/sch_hhf.c:367:23: warning: potentially expensive pointer subtraction
net/sched/sch_hhf.c:434:38: warning: potentially expensive pointer subtraction
net/sunrpc/svc_xprt.c:1377:43: warning: potentially expensive pointer subtraction
sound/pci/hda/hda_generic.c:301:20: warning: potentially expensive pointer subtraction

IOW it's not terribly noisy.  Might be an interesting idea to teach sparse to
print the type in question...  Aha - with

--- a/evaluate.c
+++ b/evaluate.c
@@ -848,7 +848,8 @@ static struct symbol *evaluate_ptr_sub(struct expression *expr)
 
                if (value & (value-1)) {
                        if (Wptr_subtraction_blows)
-                               warning(expr->pos, "potentially expensive pointer subtraction");
+                               warning(expr->pos, "[%s] potentially expensive pointer subtraction",
+                                       show_typename(lbase));
                }
 
                sub->op = '-';

we get things like
drivers/gpu/drm/i915/i915_gem_execbuffer.c:435:17: warning: [struct drm_i915_gem_exec_object2] potentially expensive pointer subtraction

OK, the top sources of that warning are:
     91 struct cpufreq_frequency_table
     36 struct Indirect				(actually, that conflates ext2/ext4/minix/sysv)
     21 struct ips_scb
     18 struct runlist_element
     13 struct zone
     13 struct vring
     11 struct usbhsh_device
     10 struct xpc_partition
      9 struct skge_element 
      9 struct lock_class
      9 struct hstate
      7 struct nvme_rdma_queue
      7 struct iso_context  
      6 struct i915_power_well
      6 struct hpet_dev 
      6 struct ext4_extent
      6 struct esas2r_target
      5 struct iio_chan_spec
      5 struct hwspinlock
      4 struct myri10ge_slice_state
      4 struct ext4_extent_idx

everything beyond that is 3 instances or less...

Powered by blists - more mailing lists