Introduce a helper to copy datagram into an iovec iterator
but also update a predefined hash. This is useful for
consumers of skb_copy_datagram_iter to also support inflight
data digest without having to finish to copy and only then
traverse the iovec and calculate the digest hash.
Acked-by: NDavid S. Miller <davem@davemloft.net>
Signed-off-by: NSagi Grimberg <sagi@lightbitslabs.com>
Signed-off-by: NChristoph Hellwig <hch@lst.de>

Allow consumers that want to use iov iterator helpers and also update
a predefined hash calculation online when copying data. This is useful
when copying incoming network buffers to a local iterator and calculate
a digest on the incoming stream. nvme-tcp host driver that will be
introduced in following patches is the first consumer via
skb_copy_and_hash_datagram_iter.
Acked-by: NDavid S. Miller <davem@davemloft.net>
Signed-off-by: NSagi Grimberg <sagi@lightbitslabs.com>
Signed-off-by: NChristoph Hellwig <hch@lst.de>

The single caller to csum_and_copy_to_iter is skb_copy_and_csum_datagram
and we are trying to unite its logic with skb_copy_datagram_iter by passing
a callback to the copy function that we want to apply. Thus, we need
to make the checksum pointer private to the function.
Acked-by: NDavid S. Miller <davem@davemloft.net>
Signed-off-by: NSagi Grimberg <sagi@lightbitslabs.com>
Signed-off-by: NChristoph Hellwig <hch@lst.de>

Between v3 [1] and v4 [2] of the blkg association series, the
association point moved from generic_make_request_checks(), which is
called after the request enters the queue, to bio_set_dev(), which is when
the bio is formed before submit_bio(). When the request_queue goes away,
the blkgs supporting the request_queue are destroyed and then the
q->root_blkg is set to %NULL.

This patch adds a %NULL check to blkg_tryget_closest() to prevent the
NPE caused by the above. It also adds a guard to see if the
request_queue is dying when creating a blkg to prevent creating a blkg
for a dead request_queue.

[1] https://lore.kernel.org/lkml/20180911184137.35897-1-dennisszhou@gmail.com/
[2] https://lore.kernel.org/lkml/20181126211946.77067-1-dennis@kernel.org/

Fixes: 5cdf2e3f ("blkcg: associate blkg when associating a device")
Reported-and-tested-by: NMing Lei <ming.lei@redhat.com>
Reviewed-by: NBart Van Assche <bvanassche@acm.org>
Signed-off-by: NDennis Zhou <dennis@kernel.org>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

Currently pblk only check the size of I/O metadata and does not take
into account if this metadata is in a separate buffer or interleaved
in a single metadata buffer.

In reality only the first scenario is supported, where second mode will
break pblk functionality during any IO operation.

This patch prevents pblk to be instantiated in case device only
supports interleaved metadata.
Reviewed-by: NJavier González <javier@cnexlabs.com>
Signed-off-by: NIgor Konopko <igor.j.konopko@intel.com>
Signed-off-by: NMatias Bjørling <mb@lightnvm.io>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

Currently lightnvm and pblk uses single DMA pool, for which the entry
size always is equal to PAGE_SIZE. The contents of each entry allocated
from the DMA pool consists of a PPA list (8bytes * 64), leaving
56bytes * 64 space for metadata. Since the metadata field can be bigger,
such as 128 bytes, the static size does not cover this use-case.

This patch adds support for I/O metadata above 56 bytes by changing DMA
pool size based on device meta size and allows pblk to use OOB metadata
>=16B.
Reviewed-by: NJavier González <javier@cnexlabs.com>
Signed-off-by: NIgor Konopko <igor.j.konopko@intel.com>
Signed-off-by: NMatias Bjørling <mb@lightnvm.io>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

The previous patches deleted all the code that needed the second value
returned from part_in_flight - now the kernel only uses the first value.

Consequently, part_in_flight (and blk_mq_in_flight) may be changed so that
it only returns one value.

This patch just refactors the code, there's no functional change.
Signed-off-by: NMikulas Patocka <mpatocka@redhat.com>
Signed-off-by: NMike Snitzer <snitzer@redhat.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

Now when part_round_stats is gone, we can switch to per-cpu in-flight
counters.

We use the local-atomic type local_t, so that if part_inc_in_flight or
part_dec_in_flight is reentrantly called from an interrupt, the value will
be correct.

The other counters could be corrupted due to reentrant interrupt, but the
corruption only results in slight counter skew - the in_flight counter
must be exact, so it needs local_t.
Signed-off-by: NMikulas Patocka <mpatocka@redhat.com>
Signed-off-by: NMike Snitzer <snitzer@redhat.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

We want to convert to per-cpu in_flight counters.

The function part_round_stats needs the in_flight counter every jiffy, it
would be too costly to sum all the percpu variables every jiffy, so it
must be deleted. part_round_stats is used to calculate two counters -
time_in_queue and io_ticks.

time_in_queue can be calculated without part_round_stats, by adding the
duration of the I/O when the I/O ends (the value is almost as exact as the
previously calculated value, except that time for in-progress I/Os is not
counted).

io_ticks can be approximated by increasing the value when I/O is started
or ended and the jiffies value has changed. If the I/Os take less than a
jiffy, the value is as exact as the previously calculated value. If the
I/Os take more than a jiffy, io_ticks can drift behind the previously
calculated value.
Signed-off-by: NMikulas Patocka <mpatocka@redhat.com>
Signed-off-by: NMike Snitzer <snitzer@redhat.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

All of part_stat_* and related methods are used with preempt disabled,
so there is no need to pass cpu around to allow of them.  Just call
smp_processor_id() as needed.
Suggested-by: NJens Axboe <axboe@kernel.dk>
Signed-off-by: NMike Snitzer <snitzer@redhat.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

This reverts commit 89c83fb5.

This should have been done as part of 2f0799a0 ("mm, thp: restore
node-local hugepage allocations").  The movement of the thp allocation
policy from alloc_pages_vma() to alloc_hugepage_direct_gfpmask() was
intended to only set __GFP_THISNODE for mempolicies that are not
MPOL_BIND whereas the revert could set this regardless of mempolicy.

While the check for MPOL_BIND between alloc_hugepage_direct_gfpmask()
and alloc_pages_vma() was racy, that has since been removed since the
revert.  What is left is the possibility to use __GFP_THISNODE in
policy_node() when it is unexpected because the special handling for
hugepages in alloc_pages_vma()  was removed as part of the consolidation.

Secondly, prior to 89c83fb5, alloc_pages_vma() implemented a somewhat
different policy for hugepage allocations, which were allocated through
alloc_hugepage_vma().  For hugepage allocations, if the allocating
process's node is in the set of allowed nodes, allocate with
__GFP_THISNODE for that node (for MPOL_PREFERRED, use that node with
__GFP_THISNODE instead).  This was changed for shmem_alloc_hugepage() to
allow fallback to other nodes in 89c83fb5 as it did for new_page() in
mm/mempolicy.c which is functionally different behavior and removes the
requirement to only allocate hugepages locally.

So this commit does a full revert of 89c83fb5 instead of the partial
revert that was done in 2f0799a0.  The result is the same thp
allocation policy for 4.20 that was in 4.19.

Fixes: 89c83fb5 ("mm, thp: consolidate THP gfp handling into alloc_hugepage_direct_gfpmask")
Fixes: 2f0799a0 ("mm, thp: restore node-local hugepage allocations")
Signed-off-by: NDavid Rientjes <rientjes@google.com>
Acked-by: NVlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Smatch generates a warning:

    drivers/scsi/scsi_lib.c:1656 scsi_mq_done() warn: test_bit() takes a bit number

The problem is that SCMD_STATE_COMPLETE is supposed to be bit number 0
and not a mask like "(1 << 0)".  It is used like this:

        if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state))

The test_and_set_bit() has a shift built in so it's a double left shift
and uses bit number 1 instead of number 0.  This bug is harmless because
it's done consistently and it doesn't clash with any other flags.

Fixes: f1342709 ("scsi: Do not rely on blk-mq for double completions")
Reviewed-by: NKeith Busch <keith.busch@intel.com>
Acked-by: NMartin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: NDan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

A controller may have an internal state that is not able to successfully
process commands for a short duration. In such states, an immediate
command requeue is expected to fail. The driver may exceed its max
retry count, which permanently ends the command in failure when the same
command would succeed after waiting for the controller to be ready.

NVMe ratified TP 4033 provides a delay hint in the completion status
code for failed commands. Implement the retry delay based on the command
completion status and the controller's requested delay.

Note that requeued commands are handled per request_queue, not per
individual request. If multiple commands fail, the controller should
consistently report the desired delay time for retryable commands in
all CQEs, otherwise the requeue list may be kicked too soon.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Reviewed-by: NSagi Grimberg <sagi@grimberg.me>
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

Technical Proposal introduces an indication for SQ flow control
disable support. Expose it since we are able to operate in this mode.
Reviewed-by: NHannes Reinecke <hare@suse.com>
Signed-off-by: NSagi Grimberg <sagi@grimberg.me>
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

Only override the allowed parts of it.
Reviewed-by: NHannes Reinecke <hare@suse.com>
Signed-off-by: NSagi Grimberg <sagi@grimberg.me>
[hch: slight tweak to the NVME_TREQ_SECURE_CHANNEL_MASK definition]
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

Technical proposal 8005 "fabrics SQ flow control" introduces a mode
where a host and controller agree to omit sq_head pointer updates
when sending nvme completions.

In case the host indicated desire to operate in this mode (connect attribute)
the controller will return back a connect completion with sq_head value
of 0xffff as indication that it will omit sq_head pointer updates.

This mode saves us an atomic update in the I/O path.
Reviewed-by: NHannes Reinecke <hare@suse.com>
[hch: suggested better implementation]
Signed-off-by: NSagi Grimberg <sagi@grimberg.me>
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

All target lldd's call the cmd receive and op completions in non-isr
thread contexts. As such the IN_ISR options are not necessary.
Remove the functionality and flags, which also removes cpu assignments
to queues.
Signed-off-by: NJames Smart <jsmart2021@gmail.com>
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

Add AEN/AER values as defined by the specification
Signed-off-by: NJay Sternberg <jay.e.sternberg@intel.com>
Reviewed-by: NSagi Grimberg <sagi@grimberg.me>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

Functions nvmet_aen_disabled and nvmet_clear_aen were using
values not bit numbers ie 1 << 9 not 9 for bit function clear_bit
and test_and_set_bit.
Signed-off-by: NJay Sternberg <jay.e.sternberg@intel.com>
Reviewed-by: NPhil Cayton <phil.cayton@intel.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

If the controller supports traffic based keep alive, we restart the keep
alive timer if any admin or io commands was completed during the kato
period.  This prevents a possible starvation of keep alive commands in
the presence of heavy traffic as in such case, we already have a health
indication from the host perspective.

Only set a comp_seen indicator in case the controller supports keep
alive to minimize the overhead for pci controllers.
Signed-off-by: NSagi Grimberg <sagi@grimberg.me>
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

We are growing more controller attributes, so use a proper enumeration
for it.  For now just add the 128-bit hostid which we support.
Reviewed-by: NChaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: NHannes Reinecke <hare@suse.com>
Signed-off-by: NSagi Grimberg <sagi@grimberg.me>
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

I was a little overzealous in removing the rcu_read_lock() call from
blkcg_bio_issue_check() and it broke blk-throttle. Put it back.

Fixes: e35403a034bf ("blkcg: associate blkg when associating a device")
Signed-off-by: NDennis Zhou <dennis@kernel.org>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

blkg reference counting now uses percpu_ref rather than atomic_t. Let's
make this consistent with css_tryget. This renames blkg_try_get to
blkg_tryget and now returns a bool rather than the blkg or %NULL.
Signed-off-by: NDennis Zhou <dennis@kernel.org>
Reviewed-by: NJosef Bacik <josef@toxicpanda.com>
Acked-by: NTejun Heo <tj@kernel.org>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

Every bio is now associated with a blkg putting blkg_get, blkg_try_get,
and blkg_put on the hot path. Switch over the refcnt in blkg to use
percpu_ref.
Signed-off-by: NDennis Zhou <dennis@kernel.org>
Acked-by: NTejun Heo <tj@kernel.org>
Reviewed-by: NJosef Bacik <josef@toxicpanda.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

Now that a bio only holds a blkg reference, so clean up is simply
putting back that reference. Remove bio_disassociate_task() as it just
calls bio_disassociate_blkg() and call the latter directly.
Signed-off-by: NDennis Zhou <dennis@kernel.org>
Acked-by: NTejun Heo <tj@kernel.org>
Reviewed-by: NJosef Bacik <josef@toxicpanda.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

The previous patch in this series removed carrying around a pointer to
the css in blkg. However, the blkg association logic still relied on
taking a reference on the css to ensure we wouldn't fail in getting a
reference for the blkg.

Here the implicit dependency on the css is removed. The association
continues to rely on the tryget logic walking up the blkg tree. This
streamlines the three ways that association can happen: normal, swap,
and writeback.
Signed-off-by: NDennis Zhou <dennis@kernel.org>
Acked-by: NTejun Heo <tj@kernel.org>
Reviewed-by: NJosef Bacik <josef@toxicpanda.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

Prior patches ensured that any bio that interacts with a request_queue
is properly associated with a blkg. This makes bio->bi_css unnecessary
as blkg maintains a reference to blkcg already.

This removes the bio field bi_css and transfers corresponding uses to
access via bi_blkg.
Signed-off-by: NDennis Zhou <dennis@kernel.org>
Reviewed-by: NJosef Bacik <josef@toxicpanda.com>
Acked-by: NTejun Heo <tj@kernel.org>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

One of the goals of this series is to remove a separate reference to
the css of the bio. This can and should be accessed via bio_blkcg(). In
this patch, wbc_init_bio() now requires a bio to have a device
associated with it.
Signed-off-by: NDennis Zhou <dennis@kernel.org>
Reviewed-by: NJosef Bacik <josef@toxicpanda.com>
Acked-by: NTejun Heo <tj@kernel.org>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

A prior patch in this series added blkg association to bios issued by
cgroups. There are two other paths that we want to attribute work back
to the appropriate cgroup: swap and writeback. Here we modify the way
swap tags bios to include the blkg. Writeback will be tackle in the next
patch.
Signed-off-by: NDennis Zhou <dennis@kernel.org>
Reviewed-by: NJosef Bacik <josef@toxicpanda.com>
Acked-by: NTejun Heo <tj@kernel.org>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

bio_issue_init among other things initializes the timestamp for an IO.
Rather than have this logic handled by policies, this consolidates it to
be on the init paths (normal, clone, bounce clone).
Signed-off-by: NDennis Zhou <dennis@kernel.org>
Acked-by: NTejun Heo <tj@kernel.org>
Reviewed-by: NLiu Bo <bo.liu@linux.alibaba.com>
Reviewed-by: NJosef Bacik <josef@toxicpanda.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

Previously, blkg association was handled by controller specific code in
blk-throttle and blk-iolatency. However, because a blkg represents a
relationship between a blkcg and a request_queue, it makes sense to keep
the blkg->q and bio->bi_disk->queue consistent.

This patch moves association into the bio_set_dev macro(). This should
cover the majority of cases where the device is set/changed keeping the
two pointers consistent. Fallback code is added to
blkcg_bio_issue_check() to catch any missing paths.
Signed-off-by: NDennis Zhou <dennis@kernel.org>
Reviewed-by: NJosef Bacik <josef@toxicpanda.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

There are 3 ways blkg association can happen: association with the
current css, with the page css (swap), or from the wbc css (writeback).

This patch handles how association is done for the first case where we
are associating bsaed on the current css. If there is already a blkg
associated, the css will be reused and association will be redone as the
request_queue may have changed.
Signed-off-by: NDennis Zhou <dennis@kernel.org>
Reviewed-by: NJosef Bacik <josef@toxicpanda.com>
Acked-by: NTejun Heo <tj@kernel.org>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

There are several scenarios where blkg_lookup_create() can fail such as
the blkcg dying, request_queue is dying, or simply being OOM. Most
handle this by simply falling back to the q->root_blkg and calling it a
day.

This patch implements the notion of closest blkg. During
blkg_lookup_create(), if it fails to create, return the closest blkg
found or the q->root_blkg. blkg_try_get_closest() is introduced and used
during association so a bio is always attached to a blkg.
Signed-off-by: NDennis Zhou <dennis@kernel.org>
Acked-by: NTejun Heo <tj@kernel.org>
Reviewed-by: NJosef Bacik <josef@toxicpanda.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

To know when to create a blkg, the general pattern is to do a
blkg_lookup() and if that fails, lock and do the lookup again, and if
that fails finally create. It doesn't make much sense for everyone who
wants to do creation to write this themselves.

This changes blkg_lookup_create() to do locking and implement this
pattern. The old blkg_lookup_create() is renamed to
__blkg_lookup_create().  If a call site wants to do its own error
handling or already owns the queue lock, they can use
__blkg_lookup_create(). This will be used in upcoming patches.
Signed-off-by: NDennis Zhou <dennis@kernel.org>
Reviewed-by: NJosef Bacik <josef@toxicpanda.com>
Acked-by: NTejun Heo <tj@kernel.org>
Reviewed-by: NLiu Bo <bo.liu@linux.alibaba.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

The bio_blkcg() function turns out to be inconsistent and consequently
dangerous to use. The first part returns a blkcg where a reference is
owned by the bio meaning it does not need to be rcu protected. However,
the third case, the last line, is problematic:

	return css_to_blkcg(task_css(current, io_cgrp_id));

This can race against task migration and the cgroup dying. It is also
semantically different as it must be called rcu protected and is
susceptible to failure when trying to get a reference to it.

This patch adds association ahead of calling bio_blkcg() rather than
after. This makes association a required and explicit step along the
code paths for calling bio_blkcg(). In blk-iolatency, association is
moved above the bio_blkcg() call to ensure it will not return %NULL.

BFQ uses the old bio_blkcg() function, but I do not want to address it
in this series due to the complexity. I have created a private version
documenting the inconsistency and noting not to use it.
Signed-off-by: NDennis Zhou <dennis@kernel.org>
Acked-by: NTejun Heo <tj@kernel.org>
Reviewed-by: NJosef Bacik <josef@toxicpanda.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

We only support polling if we have poll queues now, but the flag is
being set by default. Remove the default QUEUE_FLAG_POLL setting, we'll
set it in blk_mq_init_allocated_queue() if we have poll queues available
for this device.

Fixes: 6544d229 ("block: enable polling by default if a poll map is initalized")
Reported-by: NKirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>

While skb_push() makes the kernel panic if the skb headroom is less than
the unaligned hardware header size, it will proceed normally in case we
copy more than that because of alignment, and we'll silently corrupt
adjacent slabs.

In the case fixed by the previous patch,
"ipv6: Check available headroom in ip6_xmit() even without options", we
end up in neigh_hh_output() with 14 bytes headroom, 14 bytes hardware
header and write 16 bytes, starting 2 bytes before the allocated buffer.

Always check we're not writing before skb->head and, if the headroom is
not enough, warn and drop the packet.

v2:
 - instead of panicking with BUG_ON(), WARN_ON_ONCE() and drop the packet
   (Eric Dumazet)
 - if we avoid the panic, though, we need to explicitly check the headroom
   before the memcpy(), otherwise we'll have corrupted slabs on a running
   kernel, after we warn
 - use __skb_push() instead of skb_push(), as the headroom check is
   already implemented here explicitly (Eric Dumazet)
Signed-off-by: NStefano Brivio <sbrivio@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The broken macros make the glibc compile error. If there is no
__NR3264_fstat*, we should also removed related definitions.
Reported-by: NMarcin Juszkiewicz <marcin.juszkiewicz@linaro.org>
Fixes: bf4b6a7d ("y2038: Remove stat64 family from default syscall set")
[arnd: Both Marcin and Guo provided this patch to fix up my clearly
       broken commit, I applied the version with the better changelog.]
Signed-off-by: NGuo Ren <ren_guo@c-sky.com>
Signed-off-by: NMao Han <han_mao@c-sky.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

If for some reason an association's fragmentation point is zero,
sctp_datamsg_from_user will try to endlessly try to divide a message
into zero-sized chunks. This eventually causes kernel panic due to
running out of memory.

Although this situation is quite unlikely, it has occurred before as
reported. I propose to add this simple last-ditch sanity check due to
the severity of the potential consequences.
Signed-off-by: NJakub Audykowicz <jakub.audykowicz@gmail.com>
Acked-by: NNeil Horman <nhorman@tuxdriver.com>
Acked-by: NMarcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This is a full revert of ac5b2c18 ("mm: thp: relax __GFP_THISNODE for
MADV_HUGEPAGE mappings") and a partial revert of 89c83fb5 ("mm, thp:
consolidate THP gfp handling into alloc_hugepage_direct_gfpmask").

By not setting __GFP_THISNODE, applications can allocate remote hugepages
when the local node is fragmented or low on memory when either the thp
defrag setting is "always" or the vma has been madvised with
MADV_HUGEPAGE.

Remote access to hugepages often has much higher latency than local pages
of the native page size.  On Haswell, ac5b2c18 was shown to have a
13.9% access regression after this commit for binaries that remap their
text segment to be backed by transparent hugepages.

The intent of ac5b2c18 is to address an issue where a local node is
low on memory or fragmented such that a hugepage cannot be allocated.  In
every scenario where this was described as a fix, there is abundant and
unfragmented remote memory available to allocate from, even with a greater
access latency.

If remote memory is also low or fragmented, not setting __GFP_THISNODE was
also measured on Haswell to have a 40% regression in allocation latency.

Restore __GFP_THISNODE for thp allocations.

Fixes: ac5b2c18 ("mm: thp: relax __GFP_THISNODE for MADV_HUGEPAGE mappings")
Fixes: 89c83fb5 ("mm, thp: consolidate THP gfp handling into alloc_hugepage_direct_gfpmask")
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: NDavid Rientjes <rientjes@google.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>