Re: [PATCH -next v2] mm: annotate a data race in page_zonenum()

From: Paul E. McKenney
Date: Fri Feb 14 2020 - 11:17:45 EST


On Fri, Feb 14, 2020 at 09:30:26AM -0500, Qian Cai wrote:
> On Thu, 2020-02-13 at 11:26 -0800, Paul E. McKenney wrote:
> > On Thu, Feb 13, 2020 at 01:38:09PM -0500, Qian Cai wrote:
> > > BUG: KCSAN: data-race in page_cpupid_xchg_last / put_page
> > >
> > > write (marked) to 0xfffffc0d48ec1a00 of 8 bytes by task 91442 on cpu 3:
> > > page_cpupid_xchg_last+0x51/0x80
> > > page_cpupid_xchg_last at mm/mmzone.c:109 (discriminator 11)
> > > wp_page_reuse+0x3e/0xc0
> > > wp_page_reuse at mm/memory.c:2453
> > > do_wp_page+0x472/0x7b0
> > > do_wp_page at mm/memory.c:2798
> > > __handle_mm_fault+0xcb0/0xd00
> > > handle_pte_fault at mm/memory.c:4049
> > > (inlined by) __handle_mm_fault at mm/memory.c:4163
> > > handle_mm_fault+0xfc/0x2f0
> > > handle_mm_fault at mm/memory.c:4200
> > > do_page_fault+0x263/0x6f9
> > > do_user_addr_fault at arch/x86/mm/fault.c:1465
> > > (inlined by) do_page_fault at arch/x86/mm/fault.c:1539
> > > page_fault+0x34/0x40
> > >
> > > read to 0xfffffc0d48ec1a00 of 8 bytes by task 94817 on cpu 69:
> > > put_page+0x15a/0x1f0
> > > page_zonenum at include/linux/mm.h:923
> > > (inlined by) is_zone_device_page at include/linux/mm.h:929
> > > (inlined by) page_is_devmap_managed at include/linux/mm.h:948
> > > (inlined by) put_page at include/linux/mm.h:1023
> > > wp_page_copy+0x571/0x930
> > > wp_page_copy at mm/memory.c:2615
> > > do_wp_page+0x107/0x7b0
> > > __handle_mm_fault+0xcb0/0xd00
> > > handle_mm_fault+0xfc/0x2f0
> > > do_page_fault+0x263/0x6f9
> > > page_fault+0x34/0x40
> > >
> > > Reported by Kernel Concurrency Sanitizer on:
> > > CPU: 69 PID: 94817 Comm: systemd-udevd Tainted: G W O L 5.5.0-next-20200204+ #6
> > > Hardware name: HPE ProLiant DL385 Gen10/ProLiant DL385 Gen10, BIOS A40 07/10/2019
> > >
> > > A page never changes its zone number. The zone number happens to be
> > > stored in the same word as other bits which are modified, but the zone
> > > number bits will never be modified by any other write, so it can accept
> > > a reload of the zone bits after an intervening write and it don't need
> > > to use READ_ONCE(). Thus, annotate this data race using
> > > ASSERT_EXCLUSIVE_BITS() to also assert that there are no concurrent
> > > writes to it.
> > >
> > > Suggested-by: Marco Elver <elver@xxxxxxxxxx>
> > > Signed-off-by: Qian Cai <cai@xxxxxx>
> > > ---
> > >
> > > v2: use ASSERT_EXCLUSIVE_BITS().
> > >
> > > BTW, not sure if it is easier for Andrew with Paul to pick this up (with
> > > Andrew's ACK), since ASSERT_EXCLUSIVE_BITS() is in -rcu tree only (or likely
> > > tomorrow's -next tree).
> >
> > Here are the options I know of, any of which work for me:
> >
> > 1. I take the patch given appropriate acks/reviews.
> >
> > 2. Someone hangs onto the patch until the KCSAN infrastructure
> > hits mainline and then sends it up via whatever path.
> >
> > 3. One way to do #2 is to merge the -rcu tree's "kcsan" branch and
> > then queue this patch on top of that, again sending this patch
> > along once KCSAN hits mainline. Unusually for the -rcu tree,
> > the "kcsan" branch is not supposed to be rebased.
> >
> > Either way, just let me know!
>
> I suppose it is up to Andrew what he prefers. I may have another pile of patches
> depends on some of those KCSAN things (except the data_race() macro) are coming,
> so I'd like to find a way to get into -next first rather than waiting to send
> them in April (once KCSAN hit the mainline) if possible.

Any of the three options above would get you into -next quickly.

So just let me know how you would like to proceed.

Thanx, Paul

> > > include/linux/mm.h | 1 +
> > > 1 file changed, 1 insertion(+)
> > >
> > > diff --git a/include/linux/mm.h b/include/linux/mm.h
> > > index 52269e56c514..0d70fafd055c 100644
> > > --- a/include/linux/mm.h
> > > +++ b/include/linux/mm.h
> > > @@ -920,6 +920,7 @@ vm_fault_t alloc_set_pte(struct vm_fault *vmf, struct mem_cgroup *memcg,
> > >
> > > static inline enum zone_type page_zonenum(const struct page *page)
> > > {
> > > + ASSERT_EXCLUSIVE_BITS(page->flags, ZONES_MASK << ZONES_PGSHIFT);
> > > return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK;
> > > }
> > >
> > > --
> > > 1.8.3.1
> > >