]> git.proxmox.com Git - mirror_ubuntu-kernels.git/log
mirror_ubuntu-kernels.git
7 years agoautofs: make disc device user accessible
Ian Kent [Fri, 8 Sep 2017 23:16:27 +0000 (16:16 -0700)]
autofs: make disc device user accessible

The autofs miscellanous device ioctls that shouldn't require
CAP_SYS_ADMIN need to be accessible to user space applications in order
to be able to get information about autofs mounts.

The module checks capabilities so the miscelaneous device should be fine
with broad permissions.

Link: http://lkml.kernel.org/r/150216641928.11652.7388977863125547969.stgit@pluto.themaw.net
Signed-off-by: Ian Kent <raven@themaw.net>
Cc: Colin Walters <walters@redhat.com>
Cc: Ondrej Holy <oholy@redhat.com>
Cc: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoautofs: fix AT_NO_AUTOMOUNT not being honored
Ian Kent [Fri, 8 Sep 2017 23:16:24 +0000 (16:16 -0700)]
autofs: fix AT_NO_AUTOMOUNT not being honored

The fstatat(2) and statx() calls can pass the flag AT_NO_AUTOMOUNT which
is meant to clear the LOOKUP_AUTOMOUNT flag and prevent triggering of an
automount by the call.  But this flag is unconditionally cleared for all
stat family system calls except statx().

stat family system calls have always triggered mount requests for the
negative dentry case in follow_automount() which is intended but prevents
the fstatat(2) and statx() AT_NO_AUTOMOUNT case from being handled.

In order to handle the AT_NO_AUTOMOUNT for both system calls the negative
dentry case in follow_automount() needs to be changed to return ENOENT
when the LOOKUP_AUTOMOUNT flag is clear (and the other required flags are
clear).

AFAICT this change doesn't have any noticable side effects and may, in
some use cases (although I didn't see it in testing) prevent unnecessary
callbacks to the automount daemon.

It's also possible that a stat family call has been made with a path that
is in the process of being mounted by some other process.  But stat family
calls should return the automount state of the path as it is "now" so it
shouldn't wait for mount completion.

This is the same semantic as the positive dentry case already handled.

Link: http://lkml.kernel.org/r/150216641255.11652.4204561328197919771.stgit@pluto.themaw.net
Fixes: deccf497d804a4c5fca ("Make stat/lstat/fstatat pass AT_NO_AUTOMOUNT to vfs_statx()")
Signed-off-by: Ian Kent <raven@themaw.net>
Cc: David Howells <dhowells@redhat.com>
Cc: Colin Walters <walters@redhat.com>
Cc: Ondrej Holy <oholy@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoinit/main.c: extract early boot entropy from the passed cmdline
Daniel Micay [Fri, 8 Sep 2017 23:16:20 +0000 (16:16 -0700)]
init/main.c: extract early boot entropy from the passed cmdline

Feed the boot command-line as to the /dev/random entropy pool

Existing Android bootloaders usually pass data which may not be known by
an external attacker on the kernel command-line.  It may also be the
case on other embedded systems.  Sample command-line from a Google Pixel
running CopperheadOS....

    console=ttyHSL0,115200,n8 androidboot.console=ttyHSL0
    androidboot.hardware=sailfish user_debug=31 ehci-hcd.park=3
    lpm_levels.sleep_disabled=1 cma=32M@0-0xffffffff buildvariant=user
    veritykeyid=id:dfcb9db0089e5b3b4090a592415c28e1cb4545ab
    androidboot.bootdevice=624000.ufshc androidboot.verifiedbootstate=yellow
    androidboot.veritymode=enforcing androidboot.keymaster=1
    androidboot.serialno=FA6CE0305299 androidboot.baseband=msm
    mdss_mdp.panel=1:dsi:0:qcom,mdss_dsi_samsung_ea8064tg_1080p_cmd:1:none:cfg:single_dsi
    androidboot.slot_suffix=_b fpsimd.fpsimd_settings=0
    app_setting.use_app_setting=0 kernelflag=0x00000000 debugflag=0x00000000
    androidboot.hardware.revision=PVT radioflag=0x00000000
    radioflagex1=0x00000000 radioflagex2=0x00000000 cpumask=0x00000000
    androidboot.hardware.ddr=4096MB,Hynix,LPDDR4 androidboot.ddrinfo=00000006
    androidboot.ddrsize=4GB androidboot.hardware.color=GRA00
    androidboot.hardware.ufs=32GB,Samsung androidboot.msm.hw_ver_id=268824801
    androidboot.qf.st=2 androidboot.cid=11111111 androidboot.mid=G-2PW4100
    androidboot.bootloader=8996-012001-1704121145
    androidboot.oem_unlock_support=1 androidboot.fp_src=1
    androidboot.htc.hrdump=detected androidboot.ramdump.opt=mem@2g:2g,mem@4g:2g
    androidboot.bootreason=reboot androidboot.ramdump_enable=0 ro
    root=/dev/dm-0 dm="system none ro,0 1 android-verity /dev/sda34"
    rootwait skip_initramfs init=/init androidboot.wificountrycode=US
    androidboot.boottime=1BLL:85,1BLE:669,2BLL:0,2BLE:1777,SW:6,KL:8136

Among other things, it contains a value unique to the device
(androidboot.serialno=FA6CE0305299), unique to the OS builds for the
device variant (veritykeyid=id:dfcb9db0089e5b3b4090a592415c28e1cb4545ab)
and timings from the bootloader stages in milliseconds
(androidboot.boottime=1BLL:85,1BLE:669,2BLL:0,2BLE:1777,SW:6,KL:8136).

[tytso@mit.edu: changelog tweak]
[labbott@redhat.com: line-wrapped command line]
Link: http://lkml.kernel.org/r/20170816231458.2299-3-labbott@redhat.com
Signed-off-by: Daniel Micay <danielmicay@gmail.com>
Signed-off-by: Laura Abbott <labbott@redhat.com>
Acked-by: Kees Cook <keescook@chromium.org>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Laura Abbott <lauraa@codeaurora.org>
Cc: Nick Kralevich <nnk@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoinit: move stack canary initialization after setup_arch
Laura Abbott [Fri, 8 Sep 2017 23:16:17 +0000 (16:16 -0700)]
init: move stack canary initialization after setup_arch

Patch series "Command line randomness", v3.

A series to add the kernel command line as a source of randomness.

This patch (of 2):

Stack canary intialization involves getting a random number.  Getting this
random number may involve accessing caches or other architectural specific
features which are not available until after the architecture is setup.
Move the stack canary initialization later to accommodate this.

Link: http://lkml.kernel.org/r/20170816231458.2299-2-labbott@redhat.com
Signed-off-by: Laura Abbott <lauraa@codeaurora.org>
Signed-off-by: Laura Abbott <labbott@redhat.com>
Acked-by: Kees Cook <keescook@chromium.org>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Daniel Micay <danielmicay@gmail.com>
Cc: Nick Kralevich <nnk@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agobinfmt_flat: delete two error messages for a failed memory allocation in decompress_e...
Markus Elfring [Fri, 8 Sep 2017 23:16:14 +0000 (16:16 -0700)]
binfmt_flat: delete two error messages for a failed memory allocation in decompress_exec()

Omit extra messages for a memory allocation failure in this function.

This issue was detected by using the Coccinelle software.

Link: http://lkml.kernel.org/r/f92aac79-b05e-321a-1a19-d38c7159ee9c@users.sourceforge.net
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agocheckpatch: add 6 missing types to --list-types
Jean Delvare [Fri, 8 Sep 2017 23:16:11 +0000 (16:16 -0700)]
checkpatch: add 6 missing types to --list-types

Unlike all other types, LONG_LINE, LONG_LINE_COMMENT and LONG_LINE_STRING
are passed to WARN() through a variable.  This causes the parser in
list_types() to miss them and consequently they are not present in the
output of --list-types.

Additionally, types TYPO_SPELLING, FSF_MAILING_ADDRESS and AVOID_BUG are
passed with a variable level, causing the parser to miss them too.

So modify the regex to also catch these special cases.

Link: http://lkml.kernel.org/r/20170902175610.7e4a7c9d@endymion
Fixes: 3beb42eced39 ("checkpatch: add --list-types to show message types to show or ignore")
Signed-off-by: Jean Delvare <jdelvare@suse.de>
Acked-by: Joe Perches <joe@perches.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agocheckpatch: rename variables to avoid confusion
Jean Delvare [Fri, 8 Sep 2017 23:16:07 +0000 (16:16 -0700)]
checkpatch: rename variables to avoid confusion

The variable name "$msg_type" is sometimes used to set the message type,
and sometimes used to set the message level.  This works but is kind of
confusing.  Use "$msg_level" in the latter case instead, to make the code
clearer.

Link: http://lkml.kernel.org/r/20170902175345.175db33a@endymion
Signed-off-by: Jean Delvare <jdelvare@suse.de>
Acked-by: Joe Perches <joe@perches.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agocheckpatch: fix typo in comment
Jean Delvare [Fri, 8 Sep 2017 23:16:04 +0000 (16:16 -0700)]
checkpatch: fix typo in comment

Link: http://lkml.kernel.org/r/20170902175249.15bb77f2@endymion
Signed-off-by: Jean Delvare <jdelvare@suse.de>
Acked-by: Joe Perches <joe@perches.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agocheckpatch: add --strict check for ifs with unnecessary parentheses
Joe Perches [Fri, 8 Sep 2017 23:16:01 +0000 (16:16 -0700)]
checkpatch: add --strict check for ifs with unnecessary parentheses

An if statement test like
if ((foo == bar) && (baz != qux))
can arguably be better written without the parentheses as
if (foo == bar && baz != qux)

Add a test to find these cases.

Link: http://lkml.kernel.org/r/dcd0561ddd0fa43c51a420d53b550d738bf42001.1502734458.git.joe@perches.com
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agolib/oid_registry.c: X.509: fix the buffer overflow in the utility function for OID...
Takashi Iwai [Fri, 8 Sep 2017 23:15:58 +0000 (16:15 -0700)]
lib/oid_registry.c: X.509: fix the buffer overflow in the utility function for OID string

The sprint_oid() utility function doesn't properly check the buffer size
that it causes that the warning in vsnprintf() be triggered.  For
example on v4.1 kernel:

  ------------[ cut here ]------------
  WARNING: CPU: 0 PID: 2357 at lib/vsprintf.c:1867 vsnprintf+0x5a7/0x5c0()
  ...

We can trigger this issue by injecting maliciously crafted x509 cert in
DER format.  Just using hex editor to change the length of OID to over
the length of the SEQUENCE container.  For example:

    0:d=0  hl=4 l= 980 cons: SEQUENCE
    4:d=1  hl=4 l= 700 cons:  SEQUENCE
    8:d=2  hl=2 l=   3 cons:   cont [ 0 ]
   10:d=3  hl=2 l=   1 prim:    INTEGER           :02
   13:d=2  hl=2 l=   9 prim:   INTEGER           :9B47FAF791E7D1E3
   24:d=2  hl=2 l=  13 cons:   SEQUENCE
   26:d=3  hl=2 l=   9 prim:    OBJECT            :sha256WithRSAEncryption
   37:d=3  hl=2 l=   0 prim:    NULL
   39:d=2  hl=2 l= 121 cons:   SEQUENCE
   41:d=3  hl=2 l=  22 cons:    SET
   43:d=4  hl=2 l=  20 cons:     SEQUENCE      <=== the SEQ length is 20
   45:d=5  hl=2 l=   3 prim:      OBJECT            :organizationName
<=== the original length is 3, change the length of OID to over the length of SEQUENCE

Pawel Wieczorkiewicz reported this problem and Takashi Iwai provided
patch to fix it by checking the bufsize in sprint_oid().

Link: http://lkml.kernel.org/r/20170903021646.2080-1-jlee@suse.com
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: "Lee, Chun-Yi" <jlee@suse.com>
Reported-by: Pawel Wieczorkiewicz <pwieczorkiewicz@suse.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Pawel Wieczorkiewicz <pwieczorkiewicz@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoradix-tree: must check __radix_tree_preload() return value
Eric Dumazet [Fri, 8 Sep 2017 23:15:54 +0000 (16:15 -0700)]
radix-tree: must check __radix_tree_preload() return value

__radix_tree_preload() only disables preemption if no error is returned.

So we really need to make sure callers always check the return value.

idr_preload() contract is to always disable preemption, so we need
to add a missing preempt_disable() if an error happened.

Similarly, ida_pre_get() only needs to call preempt_enable() in the
case no error happened.

Link: http://lkml.kernel.org/r/1504637190.15310.62.camel@edumazet-glaptop3.roam.corp.google.com
Fixes: 0a835c4f090a ("Reimplement IDR and IDA using the radix tree")
Fixes: 7ad3d4d85c7a ("ida: Move ida_bitmap to a percpu variable")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: <stable@vger.kernel.org> [4.11+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agolib/cmdline.c: remove meaningless comment
Baoquan He [Fri, 8 Sep 2017 23:15:51 +0000 (16:15 -0700)]
lib/cmdline.c: remove meaningless comment

One line of code was commented out by c++ style comment for debugging, but
forgot removing it.

Clean it up.

Link: http://lkml.kernel.org/r/1503312113-11843-1-git-send-email-bhe@redhat.com
Signed-off-by: Baoquan He <bhe@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agolib/string.c: check for kmalloc() failure
Dan Carpenter [Fri, 8 Sep 2017 23:15:48 +0000 (16:15 -0700)]
lib/string.c: check for kmalloc() failure

This is mostly to keep the number of static checker warnings down so we
can spot new bugs instead of them being drowned in noise.  This function
doesn't return normal kernel error codes but instead the return value is
used to display exactly which memory failed.  I chose -1 as hopefully
that's a helpful thing to print.

Link: http://lkml.kernel.org/r/20170817115420.uikisjvfmtrqkzjn@mwanda
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Kees Cook <keescook@chromium.org>
Cc: Bjorn Helgaas <bhelgaas@google.com>
Cc: Mauro Carvalho Chehab <mchehab@kernel.org>
Cc: Heikki Krogerus <heikki.krogerus@linux.intel.com>
Cc: Daniel Micay <danielmicay@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agolib/rhashtable: fix comment on locks_mul default value
Davidlohr Bueso [Fri, 8 Sep 2017 23:15:45 +0000 (16:15 -0700)]
lib/rhashtable: fix comment on locks_mul default value

As of commit 4cf0b354d92 ("rhashtable: avoid large lock-array
allocations"), the default value for the locks multiplier was reduced
from 128 to 32.

Update the header file to reflect this.

Link: http://lkml.kernel.org/r/20170815215401.30745-1-dave@stgolabs.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Cc: Florian Westphal <fw@strlen.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agobitmap: introduce BITMAP_FROM_U64()
Yury Norov [Fri, 8 Sep 2017 23:15:41 +0000 (16:15 -0700)]
bitmap: introduce BITMAP_FROM_U64()

The macro is the compile-time analogue of bitmap_from_u64() with the same
purpose: convert the 64-bit number to the properly ordered pair of 32-bit
parts, suitable for filling the bitmap in 32-bit BE environment.

Use it to make test_bitmap_parselist() correct for 32-bit BE ABIs.

Tested on BE mips/qemu.

[akpm@linux-foundation.org: tweak code comment]
Link: http://lkml.kernel.org/r/20170810172916.24144-1-ynorov@caviumnetworks.com
Signed-off-by: Yury Norov <ynorov@caviumnetworks.com>
Cc: Noam Camus <noamca@mellanox.com>
Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Mauro Carvalho Chehab <mchehab@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agolib/test_bitmap.c: add test for bitmap_parselist()
Yury Norov [Fri, 8 Sep 2017 23:15:38 +0000 (16:15 -0700)]
lib/test_bitmap.c: add test for bitmap_parselist()

Do some basic checks for bitmap_parselist().

[akpm@linux-foundation.org: fix printk warning]
Link: http://lkml.kernel.org/r/20170807225438.16161-2-ynorov@caviumnetworks.com
Signed-off-by: Yury Norov <ynorov@caviumnetworks.com>
Cc: Noam Camus <noamca@mellanox.com>
Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Mauro Carvalho Chehab <mchehab@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agolib/bitmap.c: make bitmap_parselist() thread-safe and much faster
Yury Norov [Fri, 8 Sep 2017 23:15:34 +0000 (16:15 -0700)]
lib/bitmap.c: make bitmap_parselist() thread-safe and much faster

Current implementation of bitmap_parselist() uses a static variable to
save local state while setting bits in the bitmap.  It is obviously wrong
if we assume execution in multiprocessor environment.  Fortunately, it's
possible to rewrite this portion of code to avoid using the static
variable.

It is also possible to set bits in the mask per-range with bitmap_set(),
not per-bit, as it is implemented now, with set_bit(); which is way
faster.

The important side effect of this change is that setting bits in this
function from now is not per-bit atomic and less memory-ordered.  This is
because set_bit() guarantees the order of memory accesses, while
bitmap_set() does not.  I think that it is the advantage of the new
approach, because the bitmap_parselist() is intended to initialise bit
arrays, and user should protect the whole bitmap during initialisation if
needed.  So protecting individual bits looks expensive and useless.  Also,
other range-oriented functions in lib/bitmap.c don't worry much about
atomicity.

With all that, setting 2k bits in map with the pattern like 0-2047:128/256
becomes ~50 times faster after applying the patch in my testing
environment (arm64 hosted on qemu).

The second patch of the series adds the test for bitmap_parselist().  It's
not intended to cover all tricky cases, just to make sure that I didn't
screw up during rework.

Link: http://lkml.kernel.org/r/20170807225438.16161-1-ynorov@caviumnetworks.com
Signed-off-by: Yury Norov <ynorov@caviumnetworks.com>
Cc: Noam Camus <noamca@mellanox.com>
Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Mauro Carvalho Chehab <mchehab@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agolib: add test module for CONFIG_DEBUG_VIRTUAL
Florian Fainelli [Fri, 8 Sep 2017 23:15:31 +0000 (16:15 -0700)]
lib: add test module for CONFIG_DEBUG_VIRTUAL

Add a test module that allows testing that CONFIG_DEBUG_VIRTUAL works
correctly, at least that it can catch invalid calls to virt_to_phys()
against the non-linear kernel virtual address map.

Link: http://lkml.kernel.org/r/20170808164035.26725-1-f.fainelli@gmail.com
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Cc: "Luis R. Rodriguez" <mcgrof@kernel.org>
Cc: Kees Cook <keescook@chromium.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agolib/hexdump.c: return -EINVAL in case of error in hex2bin()
Andy Shevchenko [Fri, 8 Sep 2017 23:15:28 +0000 (16:15 -0700)]
lib/hexdump.c: return -EINVAL in case of error in hex2bin()

In some cases caller would like to use error code directly without
shadowing.

-EINVAL feels a rightful code to return in case of error in hex2bin().

Link: http://lkml.kernel.org/r/20170731135510.68023-1-andriy.shevchenko@linux.intel.com
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoblock/cfq: cache rightmost rb_node
Davidlohr Bueso [Fri, 8 Sep 2017 23:15:25 +0000 (16:15 -0700)]
block/cfq: cache rightmost rb_node

For the same reasons we already cache the leftmost pointer, apply the same
optimization for rb_last() calls.  Users must explicitly do this as
rb_root_cached only deals with the smallest node.

[dave@stgolabs.net: brain fart #1]
Link: http://lkml.kernel.org/r/20170731155955.GD21328@linux-80c1.suse
Link: http://lkml.kernel.org/r/20170719014603.19029-18-dave@stgolabs.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Cc: Jens Axboe <axboe@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomem/memcg: cache rightmost node
Davidlohr Bueso [Fri, 8 Sep 2017 23:15:21 +0000 (16:15 -0700)]
mem/memcg: cache rightmost node

Such that we can optimize __mem_cgroup_largest_soft_limit_node().  The
only overhead is the extra footprint for the cached pointer, but this
should not be an issue for mem_cgroup_tree_per_node.

[dave@stgolabs.net: brain fart #2]
Link: http://lkml.kernel.org/r/20170731160114.GE21328@linux-80c1.suse
Link: http://lkml.kernel.org/r/20170719014603.19029-17-dave@stgolabs.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agofs/epoll: use faster rb_first_cached()
Davidlohr Bueso [Fri, 8 Sep 2017 23:15:18 +0000 (16:15 -0700)]
fs/epoll: use faster rb_first_cached()

...  such that we can avoid the tree walks to get the node with the
smallest key.  Semantically the same, as the previously used rb_first(),
but O(1).  The main overhead is the extra footprint for the cached rb_node
pointer, which should not matter for epoll.

Link: http://lkml.kernel.org/r/20170719014603.19029-15-dave@stgolabs.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoprocfs: use faster rb_first_cached()
Davidlohr Bueso [Fri, 8 Sep 2017 23:15:15 +0000 (16:15 -0700)]
procfs: use faster rb_first_cached()

...  such that we can avoid the tree walks to get the node with the
smallest key.  Semantically the same, as the previously used rb_first(),
but O(1).  The main overhead is the extra footprint for the cached rb_node
pointer, which should not matter for procfs.

Link: http://lkml.kernel.org/r/20170719014603.19029-14-dave@stgolabs.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agolib/interval-tree: correct comment wrt generic flavor
Davidlohr Bueso [Fri, 8 Sep 2017 23:15:12 +0000 (16:15 -0700)]
lib/interval-tree: correct comment wrt generic flavor

interval_tree.h _is_ the generic flavor.

Link: http://lkml.kernel.org/r/20170719014603.19029-13-dave@stgolabs.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agolib/interval_tree: fast overlap detection
Davidlohr Bueso [Fri, 8 Sep 2017 23:15:08 +0000 (16:15 -0700)]
lib/interval_tree: fast overlap detection

Allow interval trees to quickly check for overlaps to avoid unnecesary
tree lookups in interval_tree_iter_first().

As of this patch, all interval tree flavors will require using a
'rb_root_cached' such that we can have the leftmost node easily
available.  While most users will make use of this feature, those with
special functions (in addition to the generic insert, delete, search
calls) will avoid using the cached option as they can do funky things
with insertions -- for example, vma_interval_tree_insert_after().

[jglisse@redhat.com: fix deadlock from typo vm_lock_anon_vma()]
Link: http://lkml.kernel.org/r/20170808225719.20723-1-jglisse@redhat.com
Link: http://lkml.kernel.org/r/20170719014603.19029-12-dave@stgolabs.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Acked-by: Christian König <christian.koenig@amd.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Doug Ledford <dledford@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Cc: David Airlie <airlied@linux.ie>
Cc: Jason Wang <jasowang@redhat.com>
Cc: Christian Benvenuti <benve@cisco.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoblock/cfq: replace cfq_rb_root leftmost caching
Davidlohr Bueso [Fri, 8 Sep 2017 23:15:05 +0000 (16:15 -0700)]
block/cfq: replace cfq_rb_root leftmost caching

... with the generic rbtree flavor instead. No changes
in semantics whatsoever.

Link: http://lkml.kernel.org/r/20170719014603.19029-11-dave@stgolabs.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Jens Axboe <axboe@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agolocking/rtmutex: replace top-waiter and pi_waiters leftmost caching
Davidlohr Bueso [Fri, 8 Sep 2017 23:15:01 +0000 (16:15 -0700)]
locking/rtmutex: replace top-waiter and pi_waiters leftmost caching

... with the generic rbtree flavor instead. No changes
in semantics whatsoever.

Link: http://lkml.kernel.org/r/20170719014603.19029-10-dave@stgolabs.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agosched/deadline: replace earliest dl and rq leftmost caching
Davidlohr Bueso [Fri, 8 Sep 2017 23:14:58 +0000 (16:14 -0700)]
sched/deadline: replace earliest dl and rq leftmost caching

... with the generic rbtree flavor instead. No changes
in semantics whatsoever.

Link: http://lkml.kernel.org/r/20170719014603.19029-9-dave@stgolabs.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agosched/fair: replace cfs_rq->rb_leftmost
Davidlohr Bueso [Fri, 8 Sep 2017 23:14:55 +0000 (16:14 -0700)]
sched/fair: replace cfs_rq->rb_leftmost

... with the generic rbtree flavor instead. No changes
in semantics whatsoever.

Link: http://lkml.kernel.org/r/20170719014603.19029-8-dave@stgolabs.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agolib/rbtree_test.c: support rb_root_cached
Davidlohr Bueso [Fri, 8 Sep 2017 23:14:52 +0000 (16:14 -0700)]
lib/rbtree_test.c: support rb_root_cached

We can work with a single rb_root_cached root to test both cached and
non-cached rbtrees.  In addition, also add a test to measure latencies
between rb_first and its fast counterpart.

Link: http://lkml.kernel.org/r/20170719014603.19029-7-dave@stgolabs.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agolib/rbtree_test.c: add (inorder) traversal test
Davidlohr Bueso [Fri, 8 Sep 2017 23:14:49 +0000 (16:14 -0700)]
lib/rbtree_test.c: add (inorder) traversal test

This adds a second test for regular rb-tree testing in that there is no
need to repeat it for the augmented flavor.

Link: http://lkml.kernel.org/r/20170719014603.19029-6-dave@stgolabs.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agolib/rbtree_test.c: make input module parameters
Davidlohr Bueso [Fri, 8 Sep 2017 23:14:46 +0000 (16:14 -0700)]
lib/rbtree_test.c: make input module parameters

Allows for more flexible debugging.

Link: http://lkml.kernel.org/r/20170719014603.19029-5-dave@stgolabs.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agorbtree: add some additional comments for rebalancing cases
Davidlohr Bueso [Fri, 8 Sep 2017 23:14:42 +0000 (16:14 -0700)]
rbtree: add some additional comments for rebalancing cases

While overall the code is very nicely commented, it might not be
immediately obvious from the diagrams what is going on.  Add a very
brief summary of each case.  Opposite cases where the node is the left
child are left untouched.

Link: http://lkml.kernel.org/r/20170719014603.19029-4-dave@stgolabs.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agorbtree: optimize root-check during rebalancing loop
Davidlohr Bueso [Fri, 8 Sep 2017 23:14:39 +0000 (16:14 -0700)]
rbtree: optimize root-check during rebalancing loop

The only times the nil-parent (root node) condition is true is when the
node is the first in the tree, or after fixing rbtree rule #4 and the
case 1 rebalancing made the node the root.  Such conditions do not apply
most of the time:

(i) The common case in an rbtree is to have more than a single node,
    so this is only true for the first rb_insert().

(ii) While there is a chance only one first rotation is needed, cases
    where the node's uncle is black (cases 2,3) are more common as we can
    have the following scenarios during the rotation looping:

    case1 only, case1+1, case2+3, case1+2+3, case3 only, etc.

This patch, therefore, adds an unlikely() optimization to this
conditional.  When profiling with CONFIG_PROFILE_ANNOTATED_BRANCHES, a
kernel build shows that the incorrect rate is less than 15%, and for
workloads that involve insert mostly trees overtime tend to have less
than 2% incorrect rate.

Link: http://lkml.kernel.org/r/20170719014603.19029-3-dave@stgolabs.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agorbtree: cache leftmost node internally
Davidlohr Bueso [Fri, 8 Sep 2017 23:14:36 +0000 (16:14 -0700)]
rbtree: cache leftmost node internally

Patch series "rbtree: Cache leftmost node internally", v4.

A series to extending rbtrees to internally cache the leftmost node such
that we can have fast overlap check optimization for all interval tree
users[1].  The benefits of this series are that:

(i)   Unify users that do internal leftmost node caching.
(ii)  Optimize all interval tree users.
(iii) Convert at least two new users (epoll and procfs) to the new interface.

This patch (of 16):

Red-black tree semantics imply that nodes with smaller or greater (or
equal for duplicates) keys always be to the left and right,
respectively.  For the kernel this is extremely evident when considering
our rb_first() semantics.  Enabling lookups for the smallest node in the
tree in O(1) can save a good chunk of cycles in not having to walk down
the tree each time.  To this end there are a few core users that
explicitly do this, such as the scheduler and rtmutexes.  There is also
the desire for interval trees to have this optimization allowing faster
overlap checking.

This patch introduces a new 'struct rb_root_cached' which is just the
root with a cached pointer to the leftmost node.  The reason why the
regular rb_root was not extended instead of adding a new structure was
that this allows the user to have the choice between memory footprint
and actual tree performance.  The new wrappers on top of the regular
rb_root calls are:

 - rb_first_cached(cached_root) -- which is a fast replacement
     for rb_first.

 - rb_insert_color_cached(node, cached_root, new)

 - rb_erase_cached(node, cached_root)

In addition, augmented cached interfaces are also added for basic
insertion and deletion operations; which becomes important for the
interval tree changes.

With the exception of the inserts, which adds a bool for updating the
new leftmost, the interfaces are kept the same.  To this end, porting rb
users to the cached version becomes really trivial, and keeping current
rbtree semantics for users that don't care about the optimization
requires zero overhead.

Link: http://lkml.kernel.org/r/20170719014603.19029-2-dave@stgolabs.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agobitops: avoid integer overflow in GENMASK(_ULL)
Matthias Kaehlcke [Fri, 8 Sep 2017 23:14:33 +0000 (16:14 -0700)]
bitops: avoid integer overflow in GENMASK(_ULL)

GENMASK(_ULL) performs a left-shift of ~0UL(L), which technically
results in an integer overflow.  clang raises a warning if the overflow
occurs in a preprocessor expression.  Clear the low-order bits through a
substraction instead of the left-shift to avoid the overflow.

(akpm: no change in .text size in my testing)

Link: http://lkml.kernel.org/r/20170803212020.24939-1-mka@chromium.org
Signed-off-by: Matthias Kaehlcke <mka@chromium.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoinclude: warn for inconsistent endian config definition
Babu Moger [Fri, 8 Sep 2017 23:14:29 +0000 (16:14 -0700)]
include: warn for inconsistent endian config definition

We have seen some generic code use config parameter CONFIG_CPU_BIG_ENDIAN
to decide the endianness.

Here are the few examples.
include/asm-generic/qrwlock.h
drivers/of/base.c
drivers/of/fdt.c
drivers/tty/serial/earlycon.c
drivers/tty/serial/serial_core.c

Display warning if CPU_BIG_ENDIAN is not defined on big endian
architecture and also warn if it defined on little endian architectures.

Here is our original discussion
https://lkml.org/lkml/2017/5/24/620

Link: http://lkml.kernel.org/r/1499358861-179979-4-git-send-email-babu.moger@oracle.com
Signed-off-by: Babu Moger <babu.moger@oracle.com>
Suggested-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: "James E.J. Bottomley" <jejb@parisc-linux.org>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: David S. Miller <davem@davemloft.net>
Cc: Greg KH <gregkh@linuxfoundation.org>
Cc: Helge Deller <deller@gmx.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Jonas Bonn <jonas@southpole.se>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Michael Ellerman <mpe@ellerman.id.au> (powerpc)
Cc: Michal Simek <monstr@monstr.eu>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stafford Horne <shorne@gmail.com>
Cc: Stefan Kristiansson <stefan.kristiansson@saunalahti.fi>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoarch/microblaze: add choice for endianness and update Makefile
Babu Moger [Fri, 8 Sep 2017 23:14:25 +0000 (16:14 -0700)]
arch/microblaze: add choice for endianness and update Makefile

microblaze architectures can be configured for either little or big endian
formats.  Add a choice option for the user to select the correct endian
format(default to big endian).

Also update the Makefile so toolchain can compile for the format it is
configured for.

Link: http://lkml.kernel.org/r/1499358861-179979-3-git-send-email-babu.moger@oracle.com
Signed-off-by: Babu Moger <babu.moger@oracle.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Michal Simek <monstr@monstr.eu>
Cc: "James E.J. Bottomley" <jejb@parisc-linux.org>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: David S. Miller <davem@davemloft.net>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Greg KH <gregkh@linuxfoundation.org>
Cc: Helge Deller <deller@gmx.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Jonas Bonn <jonas@southpole.se>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Michael Ellerman <mpe@ellerman.id.au> (powerpc)
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stafford Horne <shorne@gmail.com>
Cc: Stefan Kristiansson <stefan.kristiansson@saunalahti.fi>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoarch: define CPU_BIG_ENDIAN for all fixed big endian archs
Babu Moger [Fri, 8 Sep 2017 23:14:22 +0000 (16:14 -0700)]
arch: define CPU_BIG_ENDIAN for all fixed big endian archs

Patch series "Define CPU_BIG_ENDIAN or warn for inconsistencies", v3.

While working on enabling queued rwlock on SPARC, found this following
code in include/asm-generic/qrwlock.h which uses CONFIG_CPU_BIG_ENDIAN to
clear a byte.

static inline u8 *__qrwlock_write_byte(struct qrwlock *lock)
 {
return (u8 *)lock + 3 * IS_BUILTIN(CONFIG_CPU_BIG_ENDIAN);
 }

Problem is many of the fixed big endian architectures don't define
CPU_BIG_ENDIAN and clears the wrong byte.

Define CPU_BIG_ENDIAN for all the fixed big endian architecture to fix it.

Also found few more references of this config parameter in
drivers/of/base.c
drivers/of/fdt.c
drivers/tty/serial/earlycon.c
drivers/tty/serial/serial_core.c
Be aware that this may cause regressions if someone has worked-around
problems in the above code already. Remove the work-around.

Here is our original discussion
https://lkml.org/lkml/2017/5/24/620

Link: http://lkml.kernel.org/r/1499358861-179979-2-git-send-email-babu.moger@oracle.com
Signed-off-by: Babu Moger <babu.moger@oracle.com>
Suggested-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Stafford Horne <shorne@gmail.com>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Cc: Jonas Bonn <jonas@southpole.se>
Cc: Stefan Kristiansson <stefan.kristiansson@saunalahti.fi>
Cc: "James E.J. Bottomley" <jejb@parisc-linux.org>
Cc: Helge Deller <deller@gmx.de>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Michal Simek <monstr@monstr.eu>
Cc: Michael Ellerman <mpe@ellerman.id.au> (powerpc)
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Greg KH <gregkh@linuxfoundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agotreewide: make "nr_cpu_ids" unsigned
Alexey Dobriyan [Fri, 8 Sep 2017 23:14:18 +0000 (16:14 -0700)]
treewide: make "nr_cpu_ids" unsigned

First, number of CPUs can't be negative number.

Second, different signnnedness leads to suboptimal code in the following
cases:

1)
kmalloc(nr_cpu_ids * sizeof(X));

"int" has to be sign extended to size_t.

2)
while (loff_t *pos < nr_cpu_ids)

MOVSXD is 1 byte longed than the same MOV.

Other cases exist as well. Basically compiler is told that nr_cpu_ids
can't be negative which can't be deduced if it is "int".

Code savings on allyesconfig kernel: -3KB

add/remove: 0/0 grow/shrink: 25/264 up/down: 261/-3631 (-3370)
function                                     old     new   delta
coretemp_cpu_online                          450     512     +62
rcu_init_one                                1234    1272     +38
pci_device_probe                             374     399     +25

...

pgdat_reclaimable_pages                      628     556     -72
select_fallback_rq                           446     369     -77
task_numa_find_cpu                          1923    1807    -116

Link: http://lkml.kernel.org/r/20170819114959.GA30580@avx2
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agovga: optimise console scrolling
Matthew Wilcox [Fri, 8 Sep 2017 23:14:15 +0000 (16:14 -0700)]
vga: optimise console scrolling

Where possible, call memset16(), memmove() or memcpy() instead of using
open-coded loops.  I don't like the calling convention that uses a byte
count instead of a count of u16s, but it's a little late to change that.
Reduces code size of fbcon.o by almost 400 bytes on my laptop build.

[akpm@linux-foundation.org: fix build]
Link: http://lkml.kernel.org/r/20170720184539.31609-9-willy@infradead.org
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: David Miller <davem@davemloft.net>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "James E.J. Bottomley" <jejb@linux.vnet.ibm.com>
Cc: "Martin K. Petersen" <martin.petersen@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Russell King <rmk+kernel@armlinux.org.uk>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agodrivers/scsi/sym53c8xx_2/sym_hipd.c: convert to use memset32
Matthew Wilcox [Fri, 8 Sep 2017 23:14:11 +0000 (16:14 -0700)]
drivers/scsi/sym53c8xx_2/sym_hipd.c: convert to use memset32

memset32() can be used to initialise these three arrays.  Minor code
footprint reduction.

Link: http://lkml.kernel.org/r/20170720184539.31609-8-willy@infradead.org
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Cc: "James E.J. Bottomley" <jejb@linux.vnet.ibm.com>
Cc: "Martin K. Petersen" <martin.petersen@oracle.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: David Miller <davem@davemloft.net>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Russell King <rmk+kernel@armlinux.org.uk>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agodrivers/block/zram/zram_drv.c: convert to using memset_l
Matthew Wilcox [Fri, 8 Sep 2017 23:14:07 +0000 (16:14 -0700)]
drivers/block/zram/zram_drv.c: convert to using memset_l

zram was the motivation for creating memset_l().  Minchan Kim sees a 7%
performance improvement on x86 with 100MB of non-zero deduplicatable
data:

        perf stat -r 10 dd if=/dev/zram0 of=/dev/null

vanilla:        0.232050465 seconds time elapsed ( +-  0.51% )
memset_l: 0.217219387 seconds time elapsed ( +-  0.07% )

Link: http://lkml.kernel.org/r/20170720184539.31609-7-willy@infradead.org
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Tested-by: Minchan Kim <minchan@kernel.org>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "James E.J. Bottomley" <jejb@linux.vnet.ibm.com>
Cc: "Martin K. Petersen" <martin.petersen@oracle.com>
Cc: David Miller <davem@davemloft.net>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Russell King <rmk+kernel@armlinux.org.uk>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoalpha: add support for memset16
Matthew Wilcox [Fri, 8 Sep 2017 23:14:04 +0000 (16:14 -0700)]
alpha: add support for memset16

Alpha already had an optimised fill-memory-with-16-bit-quantity
assembler routine called memsetw().  It has a slightly different calling
convention from memset16() in that it takes a byte count, not a count of
words.  That's the same convention used by ARM's __memset routines, so
rename Alpha's routine to match and add a memset16() wrapper around it.
Then convert Alpha's scr_memsetw() to call memset16() instead of
memsetw().

Link: http://lkml.kernel.org/r/20170720184539.31609-6-willy@infradead.org
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Matt Turner <mattst88@gmail.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "James E.J. Bottomley" <jejb@linux.vnet.ibm.com>
Cc: "Martin K. Petersen" <martin.petersen@oracle.com>
Cc: David Miller <davem@davemloft.net>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Russell King <rmk+kernel@armlinux.org.uk>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoARM: implement memset32 & memset64
Matthew Wilcox [Fri, 8 Sep 2017 23:14:00 +0000 (16:14 -0700)]
ARM: implement memset32 & memset64

Reuse the existing optimised memset implementation to implement an
optimised memset32 and memset64.

Link: http://lkml.kernel.org/r/20170720184539.31609-5-willy@infradead.org
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Reviewed-by: Russell King <rmk+kernel@armlinux.org.uk>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "James E.J. Bottomley" <jejb@linux.vnet.ibm.com>
Cc: "Martin K. Petersen" <martin.petersen@oracle.com>
Cc: David Miller <davem@davemloft.net>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agox86: implement memset16, memset32 & memset64
Matthew Wilcox [Fri, 8 Sep 2017 23:13:56 +0000 (16:13 -0700)]
x86: implement memset16, memset32 & memset64

These are single instructions on x86.  There's no 64-bit instruction for
x86-32, but we don't yet have any user for memset64() on 32-bit
architectures, so don't bother to implement it.

Link: http://lkml.kernel.org/r/20170720184539.31609-4-willy@infradead.org
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "James E.J. Bottomley" <jejb@linux.vnet.ibm.com>
Cc: "Martin K. Petersen" <martin.petersen@oracle.com>
Cc: David Miller <davem@davemloft.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Russell King <rmk+kernel@armlinux.org.uk>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agolib/string.c: add testcases for memset16/32/64
Matthew Wilcox [Fri, 8 Sep 2017 23:13:52 +0000 (16:13 -0700)]
lib/string.c: add testcases for memset16/32/64

[akpm@linux-foundation.org: minor tweaks]
Link: http://lkml.kernel.org/r/20170720184539.31609-3-willy@infradead.org
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "James E.J. Bottomley" <jejb@linux.vnet.ibm.com>
Cc: "Martin K. Petersen" <martin.petersen@oracle.com>
Cc: David Miller <davem@davemloft.net>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Russell King <rmk+kernel@armlinux.org.uk>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agolib/string.c: add multibyte memset functions
Matthew Wilcox [Fri, 8 Sep 2017 23:13:48 +0000 (16:13 -0700)]
lib/string.c: add multibyte memset functions

Patch series "Multibyte memset variations", v4.

A relatively common idiom we're missing is a function to fill an area of
memory with a pattern which is larger than a single byte.  I first
noticed this with a zram patch which wanted to fill a page with an
'unsigned long' value.  There turn out to be quite a few places in the
kernel which can benefit from using an optimised function rather than a
loop; sometimes text size, sometimes speed, and sometimes both.  The
optimised PowerPC version (not included here) improves performance by
about 30% on POWER8 on just the raw memset_l().

Most of the extra lines of code come from the three testcases I added.

This patch (of 8):

memset16(), memset32() and memset64() are like memset(), but allow the
caller to fill the destination with a value larger than a single byte.
memset_l() and memset_p() allow the caller to use unsigned long and
pointer values respectively.

Link: http://lkml.kernel.org/r/20170720184539.31609-2-willy@infradead.org
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "James E.J. Bottomley" <jejb@linux.vnet.ibm.com>
Cc: "Martin K. Petersen" <martin.petersen@oracle.com>
Cc: David Miller <davem@davemloft.net>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Russell King <rmk+kernel@armlinux.org.uk>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agolinux/kernel.h: move DIV_ROUND_DOWN_ULL() macro
Masahiro Yamada [Fri, 8 Sep 2017 23:13:45 +0000 (16:13 -0700)]
linux/kernel.h: move DIV_ROUND_DOWN_ULL() macro

This macro is useful to avoid link error on 32-bit systems.

We have the same definition in two drivers, so move it to
include/linux/kernel.h

While we are here, refactor DIV_ROUND_UP_ULL() by using
DIV_ROUND_DOWN_ULL().

Link: http://lkml.kernel.org/r/1500945156-12907-1-git-send-email-yamada.masahiro@socionext.com
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Acked-by: Mark Brown <broonie@kernel.org>
Cc: Cyrille Pitchen <cyrille.pitchen@wedev4u.fr>
Cc: Jaroslav Kysela <perex@perex.cz>
Cc: Takashi Iwai <tiwai@suse.com>
Cc: Liam Girdwood <lgirdwood@gmail.com>
Cc: Boris Brezillon <boris.brezillon@free-electrons.com>
Cc: Marek Vasut <marek.vasut@gmail.com>
Cc: Brian Norris <computersforpeace@gmail.com>
Cc: Richard Weinberger <richard@nod.at>
Cc: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agofs, proc: unconditional cond_resched when reading smaps
David Rientjes [Fri, 8 Sep 2017 23:13:41 +0000 (16:13 -0700)]
fs, proc: unconditional cond_resched when reading smaps

If there are large numbers of hugepages to iterate while reading
/proc/pid/smaps, the page walk never does cond_resched().  On archs
without split pmd locks, there can be significant and observable
contention on mm->page_table_lock which cause lengthy delays without
rescheduling.

Always reschedule in smaps_pte_range() if necessary since the pagewalk
iteration can be expensive.

Link: http://lkml.kernel.org/r/alpine.DEB.2.10.1708211405520.131071@chino.kir.corp.google.com
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoproc: uninline proc_create()
Alexey Dobriyan [Fri, 8 Sep 2017 23:13:38 +0000 (16:13 -0700)]
proc: uninline proc_create()

Save some code from ~320 invocations all clearing last argument.

add/remove: 3/0 grow/shrink: 0/158 up/down: 45/-702 (-657)
function                                     old     new   delta
proc_create                                    -      17     +17
__ksymtab_proc_create                          -      16     +16
__kstrtab_proc_create                          -      12     +12
yam_init_driver                              301     298      -3

...

cifs_proc_init                               249     228     -21
via_fb_pci_probe                            2304    2280     -24

Link: http://lkml.kernel.org/r/20170819094702.GA27864@avx2
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agofs, proc: remove priv argument from is_stack
Michal Hocko [Fri, 8 Sep 2017 23:13:35 +0000 (16:13 -0700)]
fs, proc: remove priv argument from is_stack

Commit b18cb64ead40 ("fs/proc: Stop trying to report thread stacks")
removed the priv parameter user in is_stack so the argument is
redundant.  Drop it.

[arnd@arndb.de: remove unused variable]
Link: http://lkml.kernel.org/r/20170801120150.1520051-1-arnd@arndb.de
Link: http://lkml.kernel.org/r/20170728075833.7241-1-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/mempolicy.c: remove BUG_ON() checks for VMA inside mpol_misplaced()
Anshuman Khandual [Fri, 8 Sep 2017 23:13:32 +0000 (16:13 -0700)]
mm/mempolicy.c: remove BUG_ON() checks for VMA inside mpol_misplaced()

VMA and its address bounds checks are too late in this function.  They
must have been verified earlier in the page fault sequence.  Hence just
remove them.

Link: http://lkml.kernel.org/r/20170901130137.7617-1-khandual@linux.vnet.ibm.com
Signed-off-by: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Suggested-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/swapfile.c: fix swapon frontswap_map memory leak on error
David Rientjes [Fri, 8 Sep 2017 23:13:29 +0000 (16:13 -0700)]
mm/swapfile.c: fix swapon frontswap_map memory leak on error

Free frontswap_map if an error is encountered before enable_swap_info().

Signed-off-by: David Rientjes <rientjes@google.com>
Reviewed-by: "Huang, Ying" <ying.huang@intel.com>
Cc: Darrick J. Wong <darrick.wong@oracle.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: <stable@vger.kernel.org> [4.12+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: kvfree the swap cluster info if the swap file is unsatisfactory
Darrick J. Wong [Fri, 8 Sep 2017 23:13:25 +0000 (16:13 -0700)]
mm: kvfree the swap cluster info if the swap file is unsatisfactory

If initializing a small swap file fails because the swap file has a
problem (holes, etc.) then we need to free the cluster info as part of
cleanup.  Unfortunately a previous patch changed the code to use kvzalloc
but did not change all the vfree calls to use kvfree.

Found by running generic/357 from xfstests.

Link: http://lkml.kernel.org/r/20170831233515.GR3775@magnolia
Fixes: 54f180d3c181 ("mm, swap: use kvzalloc to allocate some swap data structures")
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: "Huang, Ying" <ying.huang@intel.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: <stable@vger.kernel.org> [4.12+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/page_alloc.c: apply gfp_allowed_mask before the first allocation attempt
Tetsuo Handa [Fri, 8 Sep 2017 23:13:22 +0000 (16:13 -0700)]
mm/page_alloc.c: apply gfp_allowed_mask before the first allocation attempt

We are by error initializing alloc_flags before gfp_allowed_mask is
applied.  This could cause problems after pm_restrict_gfp_mask() is called
during suspend operation.  Apply gfp_allowed_mask before initializing
alloc_flags so that the first allocation attempt uses correct flags.

Link: http://lkml.kernel.org/r/201709020016.ADJ21342.OFLJHOOSMFVtFQ@I-love.SAKURA.ne.jp
Fixes: 83d4ca8148fd9092 ("mm, page_alloc: move __GFP_HARDWALL modifications out of the fastpath")
Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agotools/testing/selftests/kcmp/kcmp_test.c: add KCMP_EPOLL_TFD testing
Cyrill Gorcunov [Fri, 8 Sep 2017 23:13:19 +0000 (16:13 -0700)]
tools/testing/selftests/kcmp/kcmp_test.c: add KCMP_EPOLL_TFD testing

KCMP's KCMP_EPOLL_TFD mode merged in commit 0791e3644e5ef2 ("kcmp: add
KCMP_EPOLL_TFD mode to compare epoll target files") we've had no selftest
for it yet (except in criu development list).  Thus add it.

Link: http://lkml.kernel.org/r/20170901151620.GK1898@uranus.lan
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Andrey Vagin <avagin@openvz.org>
Cc: Pavel Emelyanov <xemul@virtuozzo.com>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Shuah Khan <shuah@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/sparse.c: fix typo in online_mem_sections
Michal Hocko [Fri, 8 Sep 2017 23:13:15 +0000 (16:13 -0700)]
mm/sparse.c: fix typo in online_mem_sections

online_mem_sections() accidentally marks online only the first section
in the given range.  This is a typo which hasn't been noticed because I
haven't tested large 2GB blocks previously.  All users of
pfn_to_online_page would get confused on the the rest of the pfn range
in the block.

All we need to fix this is to use iterator (pfn) rather than start_pfn.

Link: http://lkml.kernel.org/r/20170904112210.3401-1-mhocko@kernel.org
Fixes: 2d070eab2e82 ("mm: consider zone which is not fully populated to have holes")
Signed-off-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/memory.c: fix mem_cgroup_oom_disable() call missing
Laurent Dufour [Fri, 8 Sep 2017 23:13:12 +0000 (16:13 -0700)]
mm/memory.c: fix mem_cgroup_oom_disable() call missing

Seen while reading the code, in handle_mm_fault(), in the case
arch_vma_access_permitted() is failing the call to
mem_cgroup_oom_disable() is not made.

To fix that, move the call to mem_cgroup_oom_enable() after calling
arch_vma_access_permitted() as it should not have entered the memcg OOM.

Link: http://lkml.kernel.org/r/1504625439-31313-1-git-send-email-ldufour@linux.vnet.ibm.com
Fixes: bae473a423f6 ("mm: introduce fault_env")
Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
Acked-by: Kirill A. Shutemov <kirill@shutemov.name>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: memcontrol: use per-cpu stocks for socket memory uncharging
Roman Gushchin [Fri, 8 Sep 2017 23:13:09 +0000 (16:13 -0700)]
mm: memcontrol: use per-cpu stocks for socket memory uncharging

We've noticed a quite noticeable performance overhead on some hosts with
significant network traffic when socket memory accounting is enabled.

Perf top shows that socket memory uncharging path is hot:
  2.13%  [kernel]                [k] page_counter_cancel
  1.14%  [kernel]                [k] __sk_mem_reduce_allocated
  1.14%  [kernel]                [k] _raw_spin_lock
  0.87%  [kernel]                [k] _raw_spin_lock_irqsave
  0.84%  [kernel]                [k] tcp_ack
  0.84%  [kernel]                [k] ixgbe_poll
  0.83%  < workload >
  0.82%  [kernel]                [k] enqueue_entity
  0.68%  [kernel]                [k] __fget
  0.68%  [kernel]                [k] tcp_delack_timer_handler
  0.67%  [kernel]                [k] __schedule
  0.60%  < workload >
  0.59%  [kernel]                [k] __inet6_lookup_established
  0.55%  [kernel]                [k] __switch_to
  0.55%  [kernel]                [k] menu_select
  0.54%  libc-2.20.so            [.] __memcpy_avx_unaligned

To address this issue, the existing per-cpu stock infrastructure can be
used.

refill_stock() can be called from mem_cgroup_uncharge_skmem() to move
charge to a per-cpu stock instead of calling atomic
page_counter_uncharge().

To prevent the uncontrolled growth of per-cpu stocks, refill_stock()
will explicitly drain the cached charge, if the cached value exceeds
CHARGE_BATCH.

This allows significantly optimize the load:
  1.21%  [kernel]                [k] _raw_spin_lock
  1.01%  [kernel]                [k] ixgbe_poll
  0.92%  [kernel]                [k] _raw_spin_lock_irqsave
  0.90%  [kernel]                [k] enqueue_entity
  0.86%  [kernel]                [k] tcp_ack
  0.85%  < workload >
  0.74%  perf-11120.map          [.] 0x000000000061bf24
  0.73%  [kernel]                [k] __schedule
  0.67%  [kernel]                [k] __fget
  0.63%  [kernel]                [k] __inet6_lookup_established
  0.62%  [kernel]                [k] menu_select
  0.59%  < workload >
  0.59%  [kernel]                [k] __switch_to
  0.57%  libc-2.20.so            [.] __memcpy_avx_unaligned

Link: http://lkml.kernel.org/r/20170829100150.4580-1-guro@fb.com
Signed-off-by: Roman Gushchin <guro@fb.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: fadvise: avoid fadvise for fs without backing device
Shakeel Butt [Fri, 8 Sep 2017 23:13:05 +0000 (16:13 -0700)]
mm: fadvise: avoid fadvise for fs without backing device

The fadvise() manpage is silent on fadvise()'s effect on memory-based
filesystems (shmem, hugetlbfs & ramfs) and pseudo file systems (procfs,
sysfs, kernfs).  The current implementaion of fadvise is mostly a noop
for such filesystems except for FADV_DONTNEED which will trigger
expensive remote LRU cache draining.  This patch makes the noop of
fadvise() on such file systems very explicit.

However this change has two side effects for ramfs and one for tmpfs.
First fadvise(FADV_DONTNEED) could remove the unmapped clean zero'ed
pages of ramfs (allocated through read, readahead & read fault) and
tmpfs (allocated through read fault).  Also fadvise(FADV_WILLNEED) could
create such clean zero'ed pages for ramfs.  This change removes those
possibilities.

One of our generic libraries does fadvise(FADV_DONTNEED).  Recently we
observed high latency in fadvise() and noticed that the users have
started using tmpfs files and the latency was due to expensive remote
LRU cache draining.  For normal tmpfs files (have data written on them),
fadvise(FADV_DONTNEED) will always trigger the unneeded remote cache
draining.

Link: http://lkml.kernel.org/r/20170818011023.181465-1-shakeelb@google.com
Signed-off-by: Shakeel Butt <shakeelb@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Greg Thelen <gthelen@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/zsmalloc.c: change stat type parameter to int
Matthias Kaehlcke [Fri, 8 Sep 2017 23:13:02 +0000 (16:13 -0700)]
mm/zsmalloc.c: change stat type parameter to int

zs_stat_inc/dec/get() uses enum zs_stat_type for the stat type, however
some callers pass an enum fullness_group value.  Change the type to int to
reflect the actual use of the functions and get rid of 'enum-conversion'
warnings

Link: http://lkml.kernel.org/r/20170731175000.56538-1-mka@chromium.org
Signed-off-by: Matthias Kaehlcke <mka@chromium.org>
Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: Doug Anderson <dianders@chromium.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/mlock.c: use page_zone() instead of page_zone_id()
Joonsoo Kim [Fri, 8 Sep 2017 23:12:59 +0000 (16:12 -0700)]
mm/mlock.c: use page_zone() instead of page_zone_id()

page_zone_id() is a specialized function to compare the zone for the pages
that are within the section range.  If the section of the pages are
different, page_zone_id() can be different even if their zone is the same.
This wrong usage doesn't cause any actual problem since
__munlock_pagevec_fill() would be called again with failed index.
However, it's better to use more appropriate function here.

Link: http://lkml.kernel.org/r/1503559211-10259-1-git-send-email-iamjoonsoo.kim@lge.com
Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: consider the number in local CPUs when reading NUMA stats
Kemi Wang [Fri, 8 Sep 2017 23:12:55 +0000 (16:12 -0700)]
mm: consider the number in local CPUs when reading NUMA stats

To avoid deviation, the per cpu number of NUMA stats in
vm_numa_stat_diff[] is included when a user *reads* the NUMA stats.

Since NUMA stats does not be read by users frequently, and kernel does not
need it to make a decision, it will not be a problem to make the readers
more expensive.

Link: http://lkml.kernel.org/r/1503568801-21305-4-git-send-email-kemi.wang@intel.com
Signed-off-by: Kemi Wang <kemi.wang@intel.com>
Reported-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Cc: Aaron Lu <aaron.lu@intel.com>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Christopher Lameter <cl@linux.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Tim Chen <tim.c.chen@intel.com>
Cc: Ying Huang <ying.huang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: update NUMA counter threshold size
Kemi Wang [Fri, 8 Sep 2017 23:12:52 +0000 (16:12 -0700)]
mm: update NUMA counter threshold size

There is significant overhead in cache bouncing caused by zone counters
(NUMA associated counters) update in parallel in multi-threaded page
allocation (suggested by Dave Hansen).

This patch updates NUMA counter threshold to a fixed size of MAX_U16 - 2,
as a small threshold greatly increases the update frequency of the global
counter from local per cpu counter(suggested by Ying Huang).

The rationality is that these statistics counters don't affect the
kernel's decision, unlike other VM counters, so it's not a problem to use
a large threshold.

With this patchset, we see 31.3% drop of CPU cycles(537-->369) for per
single page allocation and reclaim on Jesper's page_bench03 benchmark.

Benchmark provided by Jesper D Brouer(increase loop times to 10000000):
https://github.com/netoptimizer/prototype-kernel/tree/master/kernel/mm/
bench

 Threshold   CPU cycles    Throughput(88 threads)
     32          799         241760478
     64          640         301628829
     125         537         358906028 <==> system by default (base)
     256         468         412397590
     512         428         450550704
     4096        399         482520943
     20000       394         489009617
     30000       395         488017817
     65533       369(-31.3%) 521661345(+45.3%) <==> with this patchset
     N/A         342(-36.3%) 562900157(+56.8%) <==> disable zone_statistics

Link: http://lkml.kernel.org/r/1503568801-21305-3-git-send-email-kemi.wang@intel.com
Signed-off-by: Kemi Wang <kemi.wang@intel.com>
Reported-by: Jesper Dangaard Brouer <brouer@redhat.com>
Suggested-by: Dave Hansen <dave.hansen@intel.com>
Suggested-by: Ying Huang <ying.huang@intel.com>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Cc: Aaron Lu <aaron.lu@intel.com>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Christopher Lameter <cl@linux.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Tim Chen <tim.c.chen@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: change the call sites of numa statistics items
Kemi Wang [Fri, 8 Sep 2017 23:12:48 +0000 (16:12 -0700)]
mm: change the call sites of numa statistics items

Patch series "Separate NUMA statistics from zone statistics", v2.

Each page allocation updates a set of per-zone statistics with a call to
zone_statistics().  As discussed in 2017 MM summit, these are a
substantial source of overhead in the page allocator and are very rarely
consumed.  This significant overhead in cache bouncing caused by zone
counters (NUMA associated counters) update in parallel in multi-threaded
page allocation (pointed out by Dave Hansen).

A link to the MM summit slides:
  http://people.netfilter.org/hawk/presentations/MM-summit2017/MM-summit2017-JesperBrouer.pdf

To mitigate this overhead, this patchset separates NUMA statistics from
zone statistics framework, and update NUMA counter threshold to a fixed
size of MAX_U16 - 2, as a small threshold greatly increases the update
frequency of the global counter from local per cpu counter (suggested by
Ying Huang).  The rationality is that these statistics counters don't
need to be read often, unlike other VM counters, so it's not a problem
to use a large threshold and make readers more expensive.

With this patchset, we see 31.3% drop of CPU cycles(537-->369, see
below) for per single page allocation and reclaim on Jesper's
page_bench03 benchmark.  Meanwhile, this patchset keeps the same style
of virtual memory statistics with little end-user-visible effects (only
move the numa stats to show behind zone page stats, see the first patch
for details).

I did an experiment of single page allocation and reclaim concurrently
using Jesper's page_bench03 benchmark on a 2-Socket Broadwell-based
server (88 processors with 126G memory) with different size of threshold
of pcp counter.

Benchmark provided by Jesper D Brouer(increase loop times to 10000000):
  https://github.com/netoptimizer/prototype-kernel/tree/master/kernel/mm/bench

   Threshold   CPU cycles    Throughput(88 threads)
      32        799         241760478
      64        640         301628829
      125       537         358906028 <==> system by default
      256       468         412397590
      512       428         450550704
      4096      399         482520943
      20000     394         489009617
      30000     395         488017817
      65533     369(-31.3%) 521661345(+45.3%) <==> with this patchset
      N/A       342(-36.3%) 562900157(+56.8%) <==> disable zone_statistics

This patch (of 3):

In this patch, NUMA statistics is separated from zone statistics
framework, all the call sites of NUMA stats are changed to use
numa-stats-specific functions, it does not have any functionality change
except that the number of NUMA stats is shown behind zone page stats
when users *read* the zone info.

E.g. cat /proc/zoneinfo
    ***Base***                           ***With this patch***
nr_free_pages 3976                         nr_free_pages 3976
nr_zone_inactive_anon 0                    nr_zone_inactive_anon 0
nr_zone_active_anon 0                      nr_zone_active_anon 0
nr_zone_inactive_file 0                    nr_zone_inactive_file 0
nr_zone_active_file 0                      nr_zone_active_file 0
nr_zone_unevictable 0                      nr_zone_unevictable 0
nr_zone_write_pending 0                    nr_zone_write_pending 0
nr_mlock     0                             nr_mlock     0
nr_page_table_pages 0                      nr_page_table_pages 0
nr_kernel_stack 0                          nr_kernel_stack 0
nr_bounce    0                             nr_bounce    0
nr_zspages   0                             nr_zspages   0
numa_hit 0                                *nr_free_cma  0*
numa_miss 0                                numa_hit     0
numa_foreign 0                             numa_miss    0
numa_interleave 0                          numa_foreign 0
numa_local   0                             numa_interleave 0
numa_other   0                             numa_local   0
*nr_free_cma 0*                            numa_other 0
    ...                                        ...
vm stats threshold: 10                     vm stats threshold: 10
    ...                                        ...

The next patch updates the numa stats counter size and threshold.

[akpm@linux-foundation.org: coding-style fixes]
Link: http://lkml.kernel.org/r/1503568801-21305-2-git-send-email-kemi.wang@intel.com
Signed-off-by: Kemi Wang <kemi.wang@intel.com>
Reported-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christopher Lameter <cl@linux.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Ying Huang <ying.huang@intel.com>
Cc: Aaron Lu <aaron.lu@intel.com>
Cc: Tim Chen <tim.c.chen@intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/memory.c: remove reduntant check for write access
Anshuman Khandual [Fri, 8 Sep 2017 23:12:45 +0000 (16:12 -0700)]
mm/memory.c: remove reduntant check for write access

Flags argument has been copied into vmf.flags and it is not changed in
between.  Hence a single write access check can be used for both PUD and
PMD.

Link: http://lkml.kernel.org/r/20170823082839.1812-1-khandual@linux.vnet.ibm.com
Signed-off-by: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agouserfaultfd: non-cooperative: closing the uffd without triggering SIGBUS
Andrea Arcangeli [Fri, 8 Sep 2017 23:12:42 +0000 (16:12 -0700)]
userfaultfd: non-cooperative: closing the uffd without triggering SIGBUS

This is an enhancement to avoid a non cooperative userfaultfd manager
having to unregister all regions before it can close the uffd after all
userfaultfd activity completed.

The UFFDIO_UNREGISTER would serialize against the handle_userfault by
taking the mmap_sem for writing, but we can simply repeat the page fault
if we detect the uffd was closed and so the regular page fault paths
should takeover.

Link: http://lkml.kernel.org/r/20170823181227.19926-1-aarcange@redhat.com
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: remove useless vma parameter to offset_il_node
Laurent Dufour [Fri, 8 Sep 2017 23:12:39 +0000 (16:12 -0700)]
mm: remove useless vma parameter to offset_il_node

While reading the code I found that offset_il_node() has a vm_area_struct
pointer parameter which is unused.

Link: http://lkml.kernel.org/r/1502899755-23146-1-git-send-email-ldufour@linux.vnet.ibm.com
Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/hmm: fix build when HMM is disabled
Jérôme Glisse [Fri, 8 Sep 2017 23:12:35 +0000 (16:12 -0700)]
mm/hmm: fix build when HMM is disabled

Combinatorial Kconfig is painfull. Withi this patch all below combination
build.

1)

2)
CONFIG_HMM_MIRROR=y

3)
CONFIG_DEVICE_PRIVATE=y

4)
CONFIG_DEVICE_PUBLIC=y

5)
CONFIG_HMM_MIRROR=y
CONFIG_DEVICE_PUBLIC=y

6)
CONFIG_HMM_MIRROR=y
CONFIG_DEVICE_PRIVATE=y

7)
CONFIG_DEVICE_PRIVATE=y
CONFIG_DEVICE_PUBLIC=y

8)
CONFIG_HMM_MIRROR=y
CONFIG_DEVICE_PRIVATE=y
CONFIG_DEVICE_PUBLIC=y

Link: http://lkml.kernel.org/r/20170826002149.20919-1-jglisse@redhat.com
Reported-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/hmm: avoid bloating arch that do not make use of HMM
Jérôme Glisse [Fri, 8 Sep 2017 23:12:32 +0000 (16:12 -0700)]
mm/hmm: avoid bloating arch that do not make use of HMM

This moves all new code including new page migration helper behind kernel
Kconfig option so that there is no codee bloat for arch or user that do
not want to use HMM or any of its associated features.

arm allyesconfig (without all the patchset, then with and this patch):
   text    data     bss     dec     hex filename
83721896 46511131 27582964 157815991 96814b7 ../without/vmlinux
83722364 46511131 27582964 157816459 968168b vmlinux

[jglisse@redhat.com: struct hmm is only use by HMM mirror functionality]
Link: http://lkml.kernel.org/r/20170825213133.27286-1-jglisse@redhat.com
[sfr@canb.auug.org.au: fix build (arm multi_v7_defconfig)]
Link: http://lkml.kernel.org/r/20170828181849.323ab81b@canb.auug.org.au
Link: http://lkml.kernel.org/r/20170818032858.7447-1-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/hmm: add new helper to hotplug CDM memory region
Jérôme Glisse [Fri, 8 Sep 2017 23:12:28 +0000 (16:12 -0700)]
mm/hmm: add new helper to hotplug CDM memory region

Unlike unaddressable memory, coherent device memory has a real resource
associated with it on the system (as CPU can address it).  Add a new
helper to hotplug such memory within the HMM framework.

Link: http://lkml.kernel.org/r/20170817000548.32038-20-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Reviewed-by: Balbir Singh <bsingharora@gmail.com>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Evgeny Baskakov <ebaskakov@nvidia.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Mark Hairgrove <mhairgrove@nvidia.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Sherry Cheung <SCheung@nvidia.com>
Cc: Subhash Gutti <sgutti@nvidia.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Bob Liu <liubo95@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/device-public-memory: device memory cache coherent with CPU
Jérôme Glisse [Fri, 8 Sep 2017 23:12:24 +0000 (16:12 -0700)]
mm/device-public-memory: device memory cache coherent with CPU

Platform with advance system bus (like CAPI or CCIX) allow device memory
to be accessible from CPU in a cache coherent fashion.  Add a new type of
ZONE_DEVICE to represent such memory.  The use case are the same as for
the un-addressable device memory but without all the corners cases.

Link: http://lkml.kernel.org/r/20170817000548.32038-19-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Evgeny Baskakov <ebaskakov@nvidia.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Mark Hairgrove <mhairgrove@nvidia.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Sherry Cheung <SCheung@nvidia.com>
Cc: Subhash Gutti <sgutti@nvidia.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Bob Liu <liubo95@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/migrate: allow migrate_vma() to alloc new page on empty entry
Jérôme Glisse [Fri, 8 Sep 2017 23:12:21 +0000 (16:12 -0700)]
mm/migrate: allow migrate_vma() to alloc new page on empty entry

This allows callers of migrate_vma() to allocate new page for empty CPU
page table entry (pte_none or back by zero page).  This is only for
anonymous memory and it won't allow new page to be instanced if the
userfaultfd is armed.

This is useful to device driver that want to migrate a range of virtual
address and would rather allocate new memory than having to fault later
on.

Link: http://lkml.kernel.org/r/20170817000548.32038-18-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Evgeny Baskakov <ebaskakov@nvidia.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Mark Hairgrove <mhairgrove@nvidia.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Sherry Cheung <SCheung@nvidia.com>
Cc: Subhash Gutti <sgutti@nvidia.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Bob Liu <liubo95@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/migrate: support un-addressable ZONE_DEVICE page in migration
Jérôme Glisse [Fri, 8 Sep 2017 23:12:17 +0000 (16:12 -0700)]
mm/migrate: support un-addressable ZONE_DEVICE page in migration

Allow to unmap and restore special swap entry of un-addressable
ZONE_DEVICE memory.

Link: http://lkml.kernel.org/r/20170817000548.32038-17-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Evgeny Baskakov <ebaskakov@nvidia.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Mark Hairgrove <mhairgrove@nvidia.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Sherry Cheung <SCheung@nvidia.com>
Cc: Subhash Gutti <sgutti@nvidia.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Bob Liu <liubo95@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/migrate: migrate_vma() unmap page from vma while collecting pages
Jérôme Glisse [Fri, 8 Sep 2017 23:12:13 +0000 (16:12 -0700)]
mm/migrate: migrate_vma() unmap page from vma while collecting pages

Common case for migration of virtual address range is page are map only
once inside the vma in which migration is taking place.  Because we
already walk the CPU page table for that range we can directly do the
unmap there and setup special migration swap entry.

Link: http://lkml.kernel.org/r/20170817000548.32038-16-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Signed-off-by: Evgeny Baskakov <ebaskakov@nvidia.com>
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Mark Hairgrove <mhairgrove@nvidia.com>
Signed-off-by: Sherry Cheung <SCheung@nvidia.com>
Signed-off-by: Subhash Gutti <sgutti@nvidia.com>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Bob Liu <liubo95@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/migrate: new memory migration helper for use with device memory
Jérôme Glisse [Fri, 8 Sep 2017 23:12:09 +0000 (16:12 -0700)]
mm/migrate: new memory migration helper for use with device memory

This patch add a new memory migration helpers, which migrate memory
backing a range of virtual address of a process to different memory (which
can be allocated through special allocator).  It differs from numa
migration by working on a range of virtual address and thus by doing
migration in chunk that can be large enough to use DMA engine or special
copy offloading engine.

Expected users are any one with heterogeneous memory where different
memory have different characteristics (latency, bandwidth, ...).  As an
example IBM platform with CAPI bus can make use of this feature to migrate
between regular memory and CAPI device memory.  New CPU architecture with
a pool of high performance memory not manage as cache but presented as
regular memory (while being faster and with lower latency than DDR) will
also be prime user of this patch.

Migration to private device memory will be useful for device that have
large pool of such like GPU, NVidia plans to use HMM for that.

Link: http://lkml.kernel.org/r/20170817000548.32038-15-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Signed-off-by: Evgeny Baskakov <ebaskakov@nvidia.com>
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Mark Hairgrove <mhairgrove@nvidia.com>
Signed-off-by: Sherry Cheung <SCheung@nvidia.com>
Signed-off-by: Subhash Gutti <sgutti@nvidia.com>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Bob Liu <liubo95@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/migrate: new migrate mode MIGRATE_SYNC_NO_COPY
Jérôme Glisse [Fri, 8 Sep 2017 23:12:06 +0000 (16:12 -0700)]
mm/migrate: new migrate mode MIGRATE_SYNC_NO_COPY

Introduce a new migration mode that allow to offload the copy to a device
DMA engine.  This changes the workflow of migration and not all
address_space migratepage callback can support this.

This is intended to be use by migrate_vma() which itself is use for thing
like HMM (see include/linux/hmm.h).

No additional per-filesystem migratepage testing is needed.  I disables
MIGRATE_SYNC_NO_COPY in all problematic migratepage() callback and i
added comment in those to explain why (part of this patch).  The commit
message is unclear it should say that any callback that wish to support
this new mode need to be aware of the difference in the migration flow
from other mode.

Some of these callbacks do extra locking while copying (aio, zsmalloc,
balloon, ...) and for DMA to be effective you want to copy multiple
pages in one DMA operations.  But in the problematic case you can not
easily hold the extra lock accross multiple call to this callback.

Usual flow is:

For each page {
 1 - lock page
 2 - call migratepage() callback
 3 - (extra locking in some migratepage() callback)
 4 - migrate page state (freeze refcount, update page cache, buffer
     head, ...)
 5 - copy page
 6 - (unlock any extra lock of migratepage() callback)
 7 - return from migratepage() callback
 8 - unlock page
}

The new mode MIGRATE_SYNC_NO_COPY:
 1 - lock multiple pages
For each page {
 2 - call migratepage() callback
 3 - abort in all problematic migratepage() callback
 4 - migrate page state (freeze refcount, update page cache, buffer
     head, ...)
} // finished all calls to migratepage() callback
 5 - DMA copy multiple pages
 6 - unlock all the pages

To support MIGRATE_SYNC_NO_COPY in the problematic case we would need a
new callback migratepages() (for instance) that deals with multiple
pages in one transaction.

Because the problematic cases are not important for current usage I did
not wanted to complexify this patchset even more for no good reason.

Link: http://lkml.kernel.org/r/20170817000548.32038-14-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Evgeny Baskakov <ebaskakov@nvidia.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Mark Hairgrove <mhairgrove@nvidia.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Sherry Cheung <SCheung@nvidia.com>
Cc: Subhash Gutti <sgutti@nvidia.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Bob Liu <liubo95@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/hmm/devmem: dummy HMM device for ZONE_DEVICE memory
Jérôme Glisse [Fri, 8 Sep 2017 23:12:02 +0000 (16:12 -0700)]
mm/hmm/devmem: dummy HMM device for ZONE_DEVICE memory

This introduce a dummy HMM device class so device driver can use it to
create hmm_device for the sole purpose of registering device memory.  It
is useful to device driver that want to manage multiple physical device
memory under same struct device umbrella.

Link: http://lkml.kernel.org/r/20170817000548.32038-13-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Signed-off-by: Evgeny Baskakov <ebaskakov@nvidia.com>
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Mark Hairgrove <mhairgrove@nvidia.com>
Signed-off-by: Sherry Cheung <SCheung@nvidia.com>
Signed-off-by: Subhash Gutti <sgutti@nvidia.com>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Bob Liu <liubo95@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/hmm/devmem: device memory hotplug using ZONE_DEVICE
Jérôme Glisse [Fri, 8 Sep 2017 23:11:58 +0000 (16:11 -0700)]
mm/hmm/devmem: device memory hotplug using ZONE_DEVICE

This introduce a simple struct and associated helpers for device driver to
use when hotpluging un-addressable device memory as ZONE_DEVICE.  It will
find a unuse physical address range and trigger memory hotplug for it
which allocates and initialize struct page for the device memory.

Device driver should use this helper during device initialization to
hotplug the device memory.  It should only need to remove the memory once
the device is going offline (shutdown or hotremove).  There should not be
any userspace API to hotplug memory expect maybe for host device driver to
allow to add more memory to a guest device driver.

Device's memory is manage by the device driver and HMM only provides
helpers to that effect.

Link: http://lkml.kernel.org/r/20170817000548.32038-12-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Signed-off-by: Evgeny Baskakov <ebaskakov@nvidia.com>
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Mark Hairgrove <mhairgrove@nvidia.com>
Signed-off-by: Sherry Cheung <SCheung@nvidia.com>
Signed-off-by: Subhash Gutti <sgutti@nvidia.com>
Signed-off-by: Balbir Singh <bsingharora@gmail.com>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Bob Liu <liubo95@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/memcontrol: support MEMORY_DEVICE_PRIVATE
Jérôme Glisse [Fri, 8 Sep 2017 23:11:54 +0000 (16:11 -0700)]
mm/memcontrol: support MEMORY_DEVICE_PRIVATE

HMM pages (private or public device pages) are ZONE_DEVICE page and thus
need special handling when it comes to lru or refcount.  This patch make
sure that memcontrol properly handle those when it face them.  Those pages
are use like regular pages in a process address space either as anonymous
page or as file back page.  So from memcg point of view we want to handle
them like regular page for now at least.

Link: http://lkml.kernel.org/r/20170817000548.32038-11-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Acked-by: Balbir Singh <bsingharora@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Evgeny Baskakov <ebaskakov@nvidia.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Mark Hairgrove <mhairgrove@nvidia.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Sherry Cheung <SCheung@nvidia.com>
Cc: Subhash Gutti <sgutti@nvidia.com>
Cc: Bob Liu <liubo95@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/memcontrol: allow to uncharge page without using page->lru field
Jérôme Glisse [Fri, 8 Sep 2017 23:11:50 +0000 (16:11 -0700)]
mm/memcontrol: allow to uncharge page without using page->lru field

HMM pages (private or public device pages) are ZONE_DEVICE page and
thus you can not use page->lru fields of those pages. This patch
re-arrange the uncharge to allow single page to be uncharge without
modifying the lru field of the struct page.

There is no change to memcontrol logic, it is the same as it was
before this patch.

Link: http://lkml.kernel.org/r/20170817000548.32038-10-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Evgeny Baskakov <ebaskakov@nvidia.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Mark Hairgrove <mhairgrove@nvidia.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Sherry Cheung <SCheung@nvidia.com>
Cc: Subhash Gutti <sgutti@nvidia.com>
Cc: Bob Liu <liubo95@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/ZONE_DEVICE: special case put_page() for device private pages
Jérôme Glisse [Fri, 8 Sep 2017 23:11:46 +0000 (16:11 -0700)]
mm/ZONE_DEVICE: special case put_page() for device private pages

A ZONE_DEVICE page that reach a refcount of 1 is free ie no longer have
any user.  For device private pages this is important to catch and thus we
need to special case put_page() for this.

Link: http://lkml.kernel.org/r/20170817000548.32038-9-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Evgeny Baskakov <ebaskakov@nvidia.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Mark Hairgrove <mhairgrove@nvidia.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Sherry Cheung <SCheung@nvidia.com>
Cc: Subhash Gutti <sgutti@nvidia.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Bob Liu <liubo95@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/ZONE_DEVICE: new type of ZONE_DEVICE for unaddressable memory
Jérôme Glisse [Fri, 8 Sep 2017 23:11:43 +0000 (16:11 -0700)]
mm/ZONE_DEVICE: new type of ZONE_DEVICE for unaddressable memory

HMM (heterogeneous memory management) need struct page to support
migration from system main memory to device memory.  Reasons for HMM and
migration to device memory is explained with HMM core patch.

This patch deals with device memory that is un-addressable memory (ie CPU
can not access it).  Hence we do not want those struct page to be manage
like regular memory.  That is why we extend ZONE_DEVICE to support
different types of memory.

A persistent memory type is define for existing user of ZONE_DEVICE and a
new device un-addressable type is added for the un-addressable memory
type.  There is a clear separation between what is expected from each
memory type and existing user of ZONE_DEVICE are un-affected by new
requirement and new use of the un-addressable type.  All specific code
path are protect with test against the memory type.

Because memory is un-addressable we use a new special swap type for when a
page is migrated to device memory (this reduces the number of maximum swap
file).

The main two additions beside memory type to ZONE_DEVICE is two callbacks.
First one, page_free() is call whenever page refcount reach 1 (which
means the page is free as ZONE_DEVICE page never reach a refcount of 0).
This allow device driver to manage its memory and associated struct page.

The second callback page_fault() happens when there is a CPU access to an
address that is back by a device page (which are un-addressable by the
CPU).  This callback is responsible to migrate the page back to system
main memory.  Device driver can not block migration back to system memory,
HMM make sure that such page can not be pin into device memory.

If device is in some error condition and can not migrate memory back then
a CPU page fault to device memory should end with SIGBUS.

[arnd@arndb.de: fix warning]
Link: http://lkml.kernel.org/r/20170823133213.712917-1-arnd@arndb.de
Link: http://lkml.kernel.org/r/20170817000548.32038-8-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Dan Williams <dan.j.williams@intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Evgeny Baskakov <ebaskakov@nvidia.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Mark Hairgrove <mhairgrove@nvidia.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Sherry Cheung <SCheung@nvidia.com>
Cc: Subhash Gutti <sgutti@nvidia.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Bob Liu <liubo95@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/memory_hotplug: introduce add_pages
Michal Hocko [Fri, 8 Sep 2017 23:11:39 +0000 (16:11 -0700)]
mm/memory_hotplug: introduce add_pages

There are new users of memory hotplug emerging.  Some of them require
different subset of arch_add_memory.  There are some which only require
allocation of struct pages without mapping those pages to the kernel
address space.  We currently have __add_pages for that purpose.  But this
is rather lowlevel and not very suitable for the code outside of the
memory hotplug.  E.g.  x86_64 wants to update max_pfn which should be done
by the caller.  Introduce add_pages() which should care about those
details if they are needed.  Each architecture should define its
implementation and select CONFIG_ARCH_HAS_ADD_PAGES.  All others use the
currently existing __add_pages.

Link: http://lkml.kernel.org/r/20170817000548.32038-7-jglisse@redhat.com
Signed-off-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Acked-by: Balbir Singh <bsingharora@gmail.com>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Evgeny Baskakov <ebaskakov@nvidia.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Mark Hairgrove <mhairgrove@nvidia.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Sherry Cheung <SCheung@nvidia.com>
Cc: Subhash Gutti <sgutti@nvidia.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Bob Liu <liubo95@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/hmm/mirror: device page fault handler
Jérôme Glisse [Fri, 8 Sep 2017 23:11:35 +0000 (16:11 -0700)]
mm/hmm/mirror: device page fault handler

This handles page fault on behalf of device driver, unlike
handle_mm_fault() it does not trigger migration back to system memory for
device memory.

Link: http://lkml.kernel.org/r/20170817000548.32038-6-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Signed-off-by: Evgeny Baskakov <ebaskakov@nvidia.com>
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Mark Hairgrove <mhairgrove@nvidia.com>
Signed-off-by: Sherry Cheung <SCheung@nvidia.com>
Signed-off-by: Subhash Gutti <sgutti@nvidia.com>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Bob Liu <liubo95@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/hmm/mirror: helper to snapshot CPU page table
Jérôme Glisse [Fri, 8 Sep 2017 23:11:31 +0000 (16:11 -0700)]
mm/hmm/mirror: helper to snapshot CPU page table

This does not use existing page table walker because we want to share
same code for our page fault handler.

Link: http://lkml.kernel.org/r/20170817000548.32038-5-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Signed-off-by: Evgeny Baskakov <ebaskakov@nvidia.com>
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Mark Hairgrove <mhairgrove@nvidia.com>
Signed-off-by: Sherry Cheung <SCheung@nvidia.com>
Signed-off-by: Subhash Gutti <sgutti@nvidia.com>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Bob Liu <liubo95@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/hmm/mirror: mirror process address space on device with HMM helpers
Jérôme Glisse [Fri, 8 Sep 2017 23:11:27 +0000 (16:11 -0700)]
mm/hmm/mirror: mirror process address space on device with HMM helpers

This is a heterogeneous memory management (HMM) process address space
mirroring.  In a nutshell this provide an API to mirror process address
space on a device.  This boils down to keeping CPU and device page table
synchronize (we assume that both device and CPU are cache coherent like
PCIe device can be).

This patch provide a simple API for device driver to achieve address space
mirroring thus avoiding each device driver to grow its own CPU page table
walker and its own CPU page table synchronization mechanism.

This is useful for NVidia GPU >= Pascal, Mellanox IB >= mlx5 and more
hardware in the future.

[jglisse@redhat.com: fix hmm for "mmu_notifier kill invalidate_page callback"]
Link: http://lkml.kernel.org/r/20170830231955.GD9445@redhat.com
Link: http://lkml.kernel.org/r/20170817000548.32038-4-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Signed-off-by: Evgeny Baskakov <ebaskakov@nvidia.com>
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Mark Hairgrove <mhairgrove@nvidia.com>
Signed-off-by: Sherry Cheung <SCheung@nvidia.com>
Signed-off-by: Subhash Gutti <sgutti@nvidia.com>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Bob Liu <liubo95@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/hmm: heterogeneous memory management (HMM for short)
Jérôme Glisse [Fri, 8 Sep 2017 23:11:23 +0000 (16:11 -0700)]
mm/hmm: heterogeneous memory management (HMM for short)

HMM provides 3 separate types of functionality:
    - Mirroring: synchronize CPU page table and device page table
    - Device memory: allocating struct page for device memory
    - Migration: migrating regular memory to device memory

This patch introduces some common helpers and definitions to all of
those 3 functionality.

Link: http://lkml.kernel.org/r/20170817000548.32038-3-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Signed-off-by: Evgeny Baskakov <ebaskakov@nvidia.com>
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Mark Hairgrove <mhairgrove@nvidia.com>
Signed-off-by: Sherry Cheung <SCheung@nvidia.com>
Signed-off-by: Subhash Gutti <sgutti@nvidia.com>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Bob Liu <liubo95@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agohmm: heterogeneous memory management documentation
Jérôme Glisse [Fri, 8 Sep 2017 23:11:19 +0000 (16:11 -0700)]
hmm: heterogeneous memory management documentation

Patch series "HMM (Heterogeneous Memory Management)", v25.

Heterogeneous Memory Management (HMM) (description and justification)

Today device driver expose dedicated memory allocation API through their
device file, often relying on a combination of IOCTL and mmap calls.
The device can only access and use memory allocated through this API.
This effectively split the program address space into object allocated
for the device and useable by the device and other regular memory
(malloc, mmap of a file, share memory, Ã¢) only accessible by
CPU (or in a very limited way by a device by pinning memory).

Allowing different isolated component of a program to use a device thus
require duplication of the input data structure using device memory
allocator.  This is reasonable for simple data structure (array, grid,
image, Ã¢) but this get extremely complex with advance data
structure (list, tree, graph, Ã¢) that rely on a web of memory
pointers.  This is becoming a serious limitation on the kind of work
load that can be offloaded to device like GPU.

New industry standard like C++, OpenCL or CUDA are pushing to remove
this barrier.  This require a shared address space between GPU device
and CPU so that GPU can access any memory of a process (while still
obeying memory protection like read only).  This kind of feature is also
appearing in various other operating systems.

HMM is a set of helpers to facilitate several aspects of address space
sharing and device memory management.  Unlike existing sharing mechanism
that rely on pining pages use by a device, HMM relies on mmu_notifier to
propagate CPU page table update to device page table.

Duplicating CPU page table is only one aspect necessary for efficiently
using device like GPU.  GPU local memory have bandwidth in the TeraBytes/
second range but they are connected to main memory through a system bus
like PCIE that is limited to 32GigaBytes/second (PCIE 4.0 16x).  Thus it
is necessary to allow migration of process memory from main system memory
to device memory.  Issue is that on platform that only have PCIE the
device memory is not accessible by the CPU with the same properties as
main memory (cache coherency, atomic operations, ...).

To allow migration from main memory to device memory HMM provides a set of
helper to hotplug device memory as a new type of ZONE_DEVICE memory which
is un-addressable by CPU but still has struct page representing it.  This
allow most of the core kernel logic that deals with a process memory to
stay oblivious of the peculiarity of device memory.

When page backing an address of a process is migrated to device memory the
CPU page table entry is set to a new specific swap entry.  CPU access to
such address triggers a migration back to system memory, just like if the
page was swap on disk.  HMM also blocks any one from pinning a ZONE_DEVICE
page so that it can always be migrated back to system memory if CPU access
it.  Conversely HMM does not migrate to device memory any page that is pin
in system memory.

To allow efficient migration between device memory and main memory a new
migrate_vma() helpers is added with this patchset.  It allows to leverage
device DMA engine to perform the copy operation.

This feature will be use by upstream driver like nouveau mlx5 and probably
other in the future (amdgpu is next suspect in line).  We are actively
working on nouveau and mlx5 support.  To test this patchset we also worked
with NVidia close source driver team, they have more resources than us to
test this kind of infrastructure and also a bigger and better userspace
eco-system with various real industry workload they can be use to test and
profile HMM.

The expected workload is a program builds a data set on the CPU (from
disk, from network, from sensors, Ã¢).  Program uses GPU API (OpenCL,
CUDA, ...) to give hint on memory placement for the input data and also
for the output buffer.  Program call GPU API to schedule a GPU job, this
happens using device driver specific ioctl.  All this is hidden from
programmer point of view in case of C++ compiler that transparently
offload some part of a program to GPU.  Program can keep doing other stuff
on the CPU while the GPU is crunching numbers.

It is expected that CPU will not access the same data set as the GPU while
GPU is working on it, but this is not mandatory.  In fact we expect some
small memory object to be actively access by both GPU and CPU concurrently
as synchronization channel and/or for monitoring purposes.  Such object
will stay in system memory and should not be bottlenecked by system bus
bandwidth (rare write and read access from both CPU and GPU).

As we are relying on device driver API, HMM does not introduce any new
syscall nor does it modify any existing ones.  It does not change any
POSIX semantics or behaviors.  For instance the child after a fork of a
process that is using HMM will not be impacted in anyway, nor is there any
data hazard between child COW or parent COW of memory that was migrated to
device prior to fork.

HMM assume a numbers of hardware features.  Device must allow device page
table to be updated at any time (ie device job must be preemptable).
Device page table must provides memory protection such as read only.
Device must track write access (dirty bit).  Device must have a minimum
granularity that match PAGE_SIZE (ie 4k).

Reviewer (just hint):
Patch 1  HMM documentation
Patch 2  introduce core infrastructure and definition of HMM, pretty
         small patch and easy to review
Patch 3  introduce the mirror functionality of HMM, it relies on
         mmu_notifier and thus someone familiar with that part would be
         in better position to review
Patch 4  is an helper to snapshot CPU page table while synchronizing with
         concurrent page table update. Understanding mmu_notifier makes
         review easier.
Patch 5  is mostly a wrapper around handle_mm_fault()
Patch 6  add new add_pages() helper to avoid modifying each arch memory
         hot plug function
Patch 7  add a new memory type for ZONE_DEVICE and also add all the logic
         in various core mm to support this new type. Dan Williams and
         any core mm contributor are best people to review each half of
         this patchset
Patch 8  special case HMM ZONE_DEVICE pages inside put_page() Kirill and
         Dan Williams are best person to review this
Patch 9  allow to uncharge a page from memory group without using the lru
         list field of struct page (best reviewer: Johannes Weiner or
         Vladimir Davydov or Michal Hocko)
Patch 10 Add support to uncharge ZONE_DEVICE page from a memory cgroup (best
         reviewer: Johannes Weiner or Vladimir Davydov or Michal Hocko)
Patch 11 add helper to hotplug un-addressable device memory as new type
         of ZONE_DEVICE memory (new type introducted in patch 3 of this
         serie). This is boiler plate code around memory hotplug and it
         also pick a free range of physical address for the device memory.
         Note that the physical address do not point to anything (at least
         as far as the kernel knows).
Patch 12 introduce a new hmm_device class as an helper for device driver
         that want to expose multiple device memory under a common fake
         device driver. This is usefull for multi-gpu configuration.
         Anyone familiar with device driver infrastructure can review
         this. Boiler plate code really.
Patch 13 add a new migrate mode. Any one familiar with page migration is
         welcome to review.
Patch 14 introduce a new migration helper (migrate_vma()) that allow to
         migrate a range of virtual address of a process using device DMA
         engine to perform the copy. It is not limited to do copy from and
         to device but can also do copy between any kind of source and
         destination memory. Again anyone familiar with migration code
         should be able to verify the logic.
Patch 15 optimize the new migrate_vma() by unmapping pages while we are
         collecting them. This can be review by any mm folks.
Patch 16 add unaddressable memory migration to helper introduced in patch
         7, this can be review by anyone familiar with migration code
Patch 17 add a feature that allow device to allocate non-present page on
         the GPU when migrating a range of address to device memory. This
         is an helper for device driver to avoid having to first allocate
         system memory before migration to device memory
Patch 18 add a new kind of ZONE_DEVICE memory for cache coherent device
         memory (CDM)
Patch 19 add an helper to hotplug CDM memory

Previous patchset posting :
v1 http://lwn.net/Articles/597289/
v2 https://lkml.org/lkml/2014/6/12/559
v3 https://lkml.org/lkml/2014/6/13/633
v4 https://lkml.org/lkml/2014/8/29/423
v5 https://lkml.org/lkml/2014/11/3/759
v6 http://lwn.net/Articles/619737/
v7 http://lwn.net/Articles/627316/
v8 https://lwn.net/Articles/645515/
v9 https://lwn.net/Articles/651553/
v10 https://lwn.net/Articles/654430/
v11 http://www.gossamer-threads.com/lists/linux/kernel/2286424
v12 http://www.kernelhub.org/?msg=972982&p=2
v13 https://lwn.net/Articles/706856/
v14 https://lkml.org/lkml/2016/12/8/344
v15 http://www.mail-archive.com/linux-kernel@xxxxxxxxxxxxxxx/msg1304107.html
v16 http://www.spinics.net/lists/linux-mm/msg119814.html
v17 https://lkml.org/lkml/2017/1/27/847
v18 https://lkml.org/lkml/2017/3/16/596
v19 https://lkml.org/lkml/2017/4/5/831
v20 https://lwn.net/Articles/720715/
v21 https://lkml.org/lkml/2017/4/24/747
v22 http://lkml.iu.edu/hypermail/linux/kernel/1705.2/05176.html
v23 https://www.mail-archive.com/linux-kernel@vger.kernel.org/msg1404788.html
v24 https://lwn.net/Articles/726691/

This patch (of 19):

This adds documentation for HMM (Heterogeneous Memory Management).  It
presents the motivation behind it, the features necessary for it to be
useful and and gives an overview of how this is implemented.

Link: http://lkml.kernel.org/r/20170817000548.32038-2-jglisse@redhat.com
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Evgeny Baskakov <ebaskakov@nvidia.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Mark Hairgrove <mhairgrove@nvidia.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Sherry Cheung <SCheung@nvidia.com>
Cc: Subhash Gutti <sgutti@nvidia.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Bob Liu <liubo95@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: memory_hotplug: memory hotremove supports thp migration
Naoya Horiguchi [Fri, 8 Sep 2017 23:11:15 +0000 (16:11 -0700)]
mm: memory_hotplug: memory hotremove supports thp migration

This patch enables thp migration for memory hotremove.

Link: http://lkml.kernel.org/r/20170717193955.20207-11-zi.yan@sent.com
Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Zi Yan <zi.yan@cs.rutgers.edu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: migrate: move_pages() supports thp migration
Naoya Horiguchi [Fri, 8 Sep 2017 23:11:12 +0000 (16:11 -0700)]
mm: migrate: move_pages() supports thp migration

This patch enables thp migration for move_pages(2).

Link: http://lkml.kernel.org/r/20170717193955.20207-10-zi.yan@sent.com
Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Zi Yan <zi.yan@cs.rutgers.edu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: mempolicy: mbind and migrate_pages support thp migration
Naoya Horiguchi [Fri, 8 Sep 2017 23:11:08 +0000 (16:11 -0700)]
mm: mempolicy: mbind and migrate_pages support thp migration

This patch enables thp migration for mbind(2) and migrate_pages(2).

Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Zi Yan <zi.yan@cs.rutgers.edu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: soft-dirty: keep soft-dirty bits over thp migration
Naoya Horiguchi [Fri, 8 Sep 2017 23:11:04 +0000 (16:11 -0700)]
mm: soft-dirty: keep soft-dirty bits over thp migration

Soft dirty bit is designed to keep tracked over page migration.  This
patch makes it work in the same manner for thp migration too.

Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Zi Yan <zi.yan@cs.rutgers.edu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: thp: check pmd migration entry in common path
Zi Yan [Fri, 8 Sep 2017 23:11:01 +0000 (16:11 -0700)]
mm: thp: check pmd migration entry in common path

When THP migration is being used, memory management code needs to handle
pmd migration entries properly.  This patch uses !pmd_present() or
is_swap_pmd() (depending on whether pmd_none() needs separate code or
not) to check pmd migration entries at the places where a pmd entry is
present.

Since pmd-related code uses split_huge_page(), split_huge_pmd(),
pmd_trans_huge(), pmd_trans_unstable(), or
pmd_none_or_trans_huge_or_clear_bad(), this patch:

1. adds pmd migration entry split code in split_huge_pmd(),

2. takes care of pmd migration entries whenever pmd_trans_huge() is present,

3. makes pmd_none_or_trans_huge_or_clear_bad() pmd migration entry aware.

Since split_huge_page() uses split_huge_pmd() and pmd_trans_unstable()
is equivalent to pmd_none_or_trans_huge_or_clear_bad(), we do not change
them.

Until this commit, a pmd entry should be:
1. pointing to a pte page,
2. is_swap_pmd(),
3. pmd_trans_huge(),
4. pmd_devmap(), or
5. pmd_none().

Signed-off-by: Zi Yan <zi.yan@cs.rutgers.edu>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: thp: enable thp migration in generic path
Zi Yan [Fri, 8 Sep 2017 23:10:57 +0000 (16:10 -0700)]
mm: thp: enable thp migration in generic path

Add thp migration's core code, including conversions between a PMD entry
and a swap entry, setting PMD migration entry, removing PMD migration
entry, and waiting on PMD migration entries.

This patch makes it possible to support thp migration.  If you fail to
allocate a destination page as a thp, you just split the source thp as
we do now, and then enter the normal page migration.  If you succeed to
allocate destination thp, you enter thp migration.  Subsequent patches
actually enable thp migration for each caller of page migration by
allowing its get_new_page() callback to allocate thps.

[zi.yan@cs.rutgers.edu: fix gcc-4.9.0 -Wmissing-braces warning]
Link: http://lkml.kernel.org/r/A0ABA698-7486-46C3-B209-E95A9048B22C@cs.rutgers.edu
[akpm@linux-foundation.org: fix x86_64 allnoconfig warning]
Signed-off-by: Zi Yan <zi.yan@cs.rutgers.edu>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: thp: introduce CONFIG_ARCH_ENABLE_THP_MIGRATION
Naoya Horiguchi [Fri, 8 Sep 2017 23:10:53 +0000 (16:10 -0700)]
mm: thp: introduce CONFIG_ARCH_ENABLE_THP_MIGRATION

Introduce CONFIG_ARCH_ENABLE_THP_MIGRATION to limit thp migration
functionality to x86_64, which should be safer at the first step.

Link: http://lkml.kernel.org/r/20170717193955.20207-5-zi.yan@sent.com
Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Zi Yan <zi.yan@cs.rutgers.edu>
Reviewed-by: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: thp: introduce separate TTU flag for thp freezing
Naoya Horiguchi [Fri, 8 Sep 2017 23:10:49 +0000 (16:10 -0700)]
mm: thp: introduce separate TTU flag for thp freezing

TTU_MIGRATION is used to convert pte into migration entry until thp
split completes.  This behavior conflicts with thp migration added later
patches, so let's introduce a new TTU flag specifically for freezing.

try_to_unmap() is used both for thp split (via freeze_page()) and page
migration (via __unmap_and_move()).  In freeze_page(), ttu_flag given
for head page is like below (assuming anonymous thp):

    (TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS | TTU_RMAP_LOCKED | \
     TTU_MIGRATION | TTU_SPLIT_HUGE_PMD)

and ttu_flag given for tail pages is:

    (TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS | TTU_RMAP_LOCKED | \
     TTU_MIGRATION)

__unmap_and_move() calls try_to_unmap() with ttu_flag:

    (TTU_MIGRATION | TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS)

Now I'm trying to insert a branch for thp migration at the top of
try_to_unmap_one() like below

static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
                       unsigned long address, void *arg)
  {
          ...
          /* PMD-mapped THP migration entry */
          if (!pvmw.pte && (flags & TTU_MIGRATION)) {
              if (!PageAnon(page))
                  continue;

              set_pmd_migration_entry(&pvmw, page);
              continue;
          }
  ...
  }

so try_to_unmap() for tail pages called by thp split can go into thp
migration code path (which converts *pmd* into migration entry), while
the expectation is to freeze thp (which converts *pte* into migration
entry.)

I detected this failure as a "bad page state" error in a testcase where
split_huge_page() is called from queue_pages_pte_range().

Link: http://lkml.kernel.org/r/20170717193955.20207-4-zi.yan@sent.com
Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Zi Yan <zi.yan@cs.rutgers.edu>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: x86: move _PAGE_SWP_SOFT_DIRTY from bit 7 to bit 1
Naoya Horiguchi [Fri, 8 Sep 2017 23:10:46 +0000 (16:10 -0700)]
mm: x86: move _PAGE_SWP_SOFT_DIRTY from bit 7 to bit 1

_PAGE_PSE is used to distinguish between a truly non-present
(_PAGE_PRESENT=0) PMD, and a PMD which is undergoing a THP split and
should be treated as present.

But _PAGE_SWP_SOFT_DIRTY currently uses the _PAGE_PSE bit, which would
cause confusion between one of those PMDs undergoing a THP split, and a
soft-dirty PMD.  Dropping _PAGE_PSE check in pmd_present() does not work
well, because it can hurt optimization of tlb handling in thp split.

Thus, we need to move the bit.

In the current kernel, bits 1-4 are not used in non-present format since
commit 00839ee3b299 ("x86/mm: Move swap offset/type up in PTE to work
around erratum").  So let's move _PAGE_SWP_SOFT_DIRTY to bit 1.  Bit 7
is used as reserved (always clear), so please don't use it for other
purpose.

Link: http://lkml.kernel.org/r/20170717193955.20207-3-zi.yan@sent.com
Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Zi Yan <zi.yan@cs.rutgers.edu>
Acked-by: Dave Hansen <dave.hansen@intel.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: mempolicy: add queue_pages_required()
Naoya Horiguchi [Fri, 8 Sep 2017 23:10:42 +0000 (16:10 -0700)]
mm: mempolicy: add queue_pages_required()

Patch series "mm: page migration enhancement for thp", v9.

Motivations:

1. THP migration becomes important in the upcoming heterogeneous memory
   systems. As David Nellans from NVIDIA pointed out from other threads
   (http://www.mail-archive.com/linux-kernel@vger.kernel.org/msg1349227.html),
   future GPUs or other accelerators will have their memory managed by
   operating systems. Moving data into and out of these memory nodes
   efficiently is critical to applications that use GPUs or other
   accelerators. Existing page migration only supports base pages, which
   has a very low memory bandwidth utilization. My experiments (see
   below) show THP migration can migrate pages more efficiently.

2. Base page migration vs THP migration throughput.

   Here are cross-socket page migration results from calling
   move_pages() syscall:

   In x86_64, a Intel two-socket E5-2640v3 box,
    - single 4KB base page migration takes 62.47 us, using 0.06 GB/s BW,
    - single 2MB THP migration takes 658.54 us, using 2.97 GB/s BW,
    - 512 4KB base page migration takes 1987.38 us, using 0.98 GB/s BW.

   In ppc64, a two-socket Power8 box,
    - single 64KB base page migration takes 49.3 us, using 1.24 GB/s BW,
    - single 16MB THP migration takes 2202.17 us, using 7.10 GB/s BW,
    - 256 64KB base page migration takes 2543.65 us, using 6.14 GB/s BW.

   THP migration can give us 3x and 1.15x throughput over base page
   migration in x86_64 and ppc64 respectivley.

   You can test it out by using the code here:
      https://github.com/x-y-z/thp-migration-bench

3. Existing page migration splits THP before migration and cannot
   guarantee the migrated pages are still contiguous. Contiguity is
   always what GPUs and accelerators look for. Without THP migration,
   khugepaged needs to do extra work to reassemble the migrated pages
   back to THPs.

This patch (of 10):

Introduce a separate check routine related to MPOL_MF_INVERT flag.  This
patch just does cleanup, no behavioral change.

Link: http://lkml.kernel.org/r/20170717193955.20207-2-zi.yan@sent.com
Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Zi Yan <zi.yan@cs.rutgers.edu>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: David Nellans <dnellans@nvidia.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>