Sorry to dig up an old thread but a coworker was asking about this
patch. This is essentially the code that landed in commit
f2f43e566a02a3bdde0a65e6a2e88d707c212a29 "mm/vmscan.c: fix unsequenced
modification and access warning".
Is .reclaim_idx still correct in the case of try_to_free_pages()? It
looks like reclaim_idx is based on the original gfp_mask in
__node_reclaim(), but in try_to_free_pages() it looks like it may have
been based on current_gfp_context()? (The sequencing is kind of
ambiguous, thus fixed in my patch)
Was there a bug in the original try_to_free_pages() pre commit
f2f43e566a0, or is .reclaim_idx supposed to be different between
try_to_free_pages() and __node_reclaim()?
On Wed, May 10, 2017 at 12:15 AM, Michal Hocko <[email protected]> wrote:
> On Tue 09-05-17 23:53:28, Nick Desaulniers wrote:
>> Clang flags this file with the -Wunsequenced error that GCC does not
>> have.
>>
>> unsequenced modification and access to 'gfp_mask'
>>
>> It seems that gfp_mask is both read and written without a sequence point
>> in between, which is undefined behavior.
>
> Hmm. This is rather news to me. I thought that a = foo(a) is perfectly
> valid. Same as a = b = c where c = foo(b) or is the problem in the
> following .reclaim_idx = gfp_zone(gfp_mask) initialization? If that is
> the case then the current code is OKish because gfp_zone doesn't depend
> on the gfp_mask modification. It is messy, right, but works as expected.
>
> Anyway, we have a similar construct __node_reclaim
>
> If you really want to change this code, and I would agree it would be
> slightly less tricky, then I would suggest doing something like the
> following instead
> ---
> diff --git a/mm/vmscan.c b/mm/vmscan.c
> index 5ebf468c5429..ba4b695e810e 100644
> --- a/mm/vmscan.c
> +++ b/mm/vmscan.c
> @@ -2965,7 +2965,7 @@ unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
> unsigned long nr_reclaimed;
> struct scan_control sc = {
> .nr_to_reclaim = SWAP_CLUSTER_MAX,
> - .gfp_mask = (gfp_mask = current_gfp_context(gfp_mask)),
> + .gfp_mask = current_gfp_context(gfp_mask),
> .reclaim_idx = gfp_zone(gfp_mask),
> .order = order,
> .nodemask = nodemask,
> @@ -2980,12 +2980,12 @@ unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
> * 1 is returned so that the page allocator does not OOM kill at this
> * point.
> */
> - if (throttle_direct_reclaim(gfp_mask, zonelist, nodemask))
> + if (throttle_direct_reclaim(sc.gfp_mask, zonelist, nodemask))
> return 1;
>
> trace_mm_vmscan_direct_reclaim_begin(order,
> sc.may_writepage,
> - gfp_mask,
> + sc.gfp_mask,
> sc.reclaim_idx);
>
> nr_reclaimed = do_try_to_free_pages(zonelist, &sc);
> @@ -3772,17 +3772,16 @@ static int __node_reclaim(struct pglist_data *pgdat, gfp_t gfp_mask, unsigned in
> const unsigned long nr_pages = 1 << order;
> struct task_struct *p = current;
> struct reclaim_state reclaim_state;
> - int classzone_idx = gfp_zone(gfp_mask);
> unsigned int noreclaim_flag;
> struct scan_control sc = {
> .nr_to_reclaim = max(nr_pages, SWAP_CLUSTER_MAX),
> - .gfp_mask = (gfp_mask = current_gfp_context(gfp_mask)),
> + .gfp_mask = current_gfp_context(gfp_mask),
> .order = order,
> .priority = NODE_RECLAIM_PRIORITY,
> .may_writepage = !!(node_reclaim_mode & RECLAIM_WRITE),
> .may_unmap = !!(node_reclaim_mode & RECLAIM_UNMAP),
> .may_swap = 1,
> - .reclaim_idx = classzone_idx,
> + .reclaim_idx = gfp_znoe(gfp_mask),
> };
>
> cond_resched();
> @@ -3793,7 +3792,7 @@ static int __node_reclaim(struct pglist_data *pgdat, gfp_t gfp_mask, unsigned in
> */
> noreclaim_flag = memalloc_noreclaim_save();
> p->flags |= PF_SWAPWRITE;
> - lockdep_set_current_reclaim_state(gfp_mask);
> + lockdep_set_current_reclaim_state(sc.gfp_mask);
> reclaim_state.reclaimed_slab = 0;
> p->reclaim_state = &reclaim_state;
>
> --
> Michal Hocko
> SUSE Labs
On Wed 21-03-18 14:37:04, Nick Desaulniers wrote:
> Sorry to dig up an old thread but a coworker was asking about this
> patch. This is essentially the code that landed in commit
> f2f43e566a02a3bdde0a65e6a2e88d707c212a29 "mm/vmscan.c: fix unsequenced
> modification and access warning".
>
> Is .reclaim_idx still correct in the case of try_to_free_pages()?
Yes, it gets initialized from the given gfp_mask. sc.gfp_mask might be
sllightly different but that doesn't change the reclaim_idx because we
only drop __GFP_{FS,IO} which do not have any zone modification effects.
> It
> looks like reclaim_idx is based on the original gfp_mask in
> __node_reclaim(), but in try_to_free_pages() it looks like it may have
> been based on current_gfp_context()? (The sequencing is kind of
> ambiguous, thus fixed in my patch)
>
> Was there a bug in the original try_to_free_pages() pre commit
> f2f43e566a0, or is .reclaim_idx supposed to be different between
> try_to_free_pages() and __node_reclaim()?
I do not think there was any real bug.
--
Michal Hocko
SUSE Labs