It needs to add warning to update firmware in gfx9
in case that firmware is too old to have function to
realize dummy read in cp firmware.
Signed-off-by: changzhu <Changfeng.Zhu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The GRBM register interface is now capable of bursting 1 cycle per
register wr->wr, wr->rd much faster than previous muticycle per
transaction done interface. This has caused a problem where
status registers requiring HW to update have a 1 cycle delay, due
to the register update having to go through GRBM.
For cp ucode, it has realized dummy read in cp firmware.It covers
the use of WAIT_REG_MEM operation 1 case only.So it needs to call
gfx_v10_0_wait_reg_mem in gfx10. Besides it also needs to add warning to
update firmware in case firmware is too old to have function to realize
dummy read in cp firmware.
For sdma ucode, it hasn't realized dummy read in sdma firmware. sdma is
moved to gfxhub in gfx10. So it needs to add dummy read in driver
between amdgpu_ring_emit_wreg and amdgpu_ring_emit_reg_wait for sdma_v5_0.
Signed-off-by: changzhu <Changfeng.Zhu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Otherwise, the feature enablement will be skipped due to wrong count.
Fixes: beff74bc6e ("drm/amdgpu: fix a race in GPU reset with IB test (v2)")
Signed-off-by: Evan Quan <evan.quan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
fix workload bit (WORKLOAD_PPLIB_COMPUTE_BIT) map error
on vega20 and navi asic.
fix commit:
drm/amd/powerplay: add function get_workload_type_map for swsmu
Signed-off-by: Kevin Wang <kevin1.wang@amd.com>
Reviewed-by: Kenneth Feng <kenneth.feng@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
If a pnet table entry is to be added mentioning a valid ethernet
interface, but an invalid infiniband or ISM device, the dev_put()
operation for the ethernet interface is called twice, resulting
in a negative refcount for the ethernet interface, which disables
removal of such a network interface.
This patch removes one of the dev_put() calls.
Fixes: 890a2cb4a9 ("net/smc: rework pnet table")
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Jakub Kicinski says:
====================
net/tls: add a TX lock
Some time ago Pooja and Mallesham started reporting crashes with
an async accelerator. After trying to poke the existing logic into
shape I came to the conclusion that it can't be trusted, and to
preserve our sanity we should just add a lock around the TX side.
First patch removes the sk_write_pending checks from the write
space callbacks. Those don't seem to have a logical justification.
Patch 2 adds the TX lock and patch 3 associated test (which should
hang with current net).
Mallesham reports that even with these fixes applied the async
accelerator workload still occasionally hangs waiting for socket
memory. I suspect that's strictly related to the way async crypto
is integrated in TLS, so I think we should get these into net or
net-next and move from there.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a test which spawns 16 threads and performs concurrent
send and recv calls on the same socket.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
TLS TX needs to release and re-acquire the socket lock if send buffer
fills up.
TLS SW TX path currently depends on only allowing one thread to enter
the function by the abuse of sk_write_pending. If another writer is
already waiting for memory no new ones are allowed in.
This has two problems:
- writers don't wake other threads up when they leave the kernel;
meaning that this scheme works for single extra thread (second
application thread or delayed work) because memory becoming
available will send a wake up request, but as Mallesham and
Pooja report with larger number of threads it leads to threads
being put to sleep indefinitely;
- the delayed work does not get _scheduled_ but it may _run_ when
other writers are present leading to crashes as writers don't
expect state to change under their feet (same records get pushed
and freed multiple times); it's hard to reliably bail from the
work, however, because the mere presence of a writer does not
guarantee that the writer will push pending records before exiting.
Ensuring wakeups always happen will make the code basically open
code a mutex. Just use a mutex.
The TLS HW TX path does not have any locking (not even the
sk_write_pending hack), yet it uses a per-socket sg_tx_data
array to push records.
Fixes: a42055e8d2 ("net/tls: Add support for async encryption of records for performance")
Reported-by: Mallesham Jatharakonda <mallesh537@gmail.com>
Reported-by: Pooja Trivedi <poojatrivedi@gmail.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
sk_write_pending being not zero does not guarantee that partial
record will be pushed. If the thread waiting for memory times out
the pending record may get stuck.
In case of tls_device there is no path where parial record is
set and writer present in the first place. Partial record is
set only in tls_push_sg() and tls_push_sg() will return an
error immediately. All tls_device callers of tls_push_sg()
will return (and not wait for memory) if it failed.
Fixes: a42055e8d2 ("net/tls: Add support for async encryption of records for performance")
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
blkcg_print_stat() iterates blkgs under RCU and doesn't test whether
the blkg is online. This can call into pd_stat_fn() on a pd which is
still being initialized leading to an oops.
The heaviest operation - recursively summing up rwstat counters - is
already done while holding the queue_lock. Expand queue_lock to cover
the other operations and skip the blkg if it isn't online yet. The
online state is protected by both blkcg and queue locks, so this
guarantees that only online blkgs are processed.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Roman Gushchin <guro@fb.com>
Cc: Josef Bacik <jbacik@fb.com>
Fixes: 903d23f0a3 ("blk-cgroup: allow controllers to output their own stats")
Cc: stable@vger.kernel.org # v4.19+
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add missing docbook comments to madvise fields in struct
drm_gem_shmem_object which fixes these warnings:
include/drm/drm_gem_shmem_helper.h:87: warning: Function parameter or member 'madv' not described in 'drm_gem_shmem_object'
include/drm/drm_gem_shmem_helper.h:87: warning: Function parameter or member 'madv_list' not described in 'drm_gem_shmem_object'
Fixes: 17acb9f35e ("drm/shmem: Add madvise state and purge helpers")
Reported-by: Sean Paul <sean@poorly.run>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Maxime Ripard <mripard@kernel.org>
Cc: David Airlie <airlied@linux.ie>
Cc: Daniel Vetter <daniel@ffwll.ch>
Signed-off-by: Rob Herring <robh@kernel.org>
Reviewed-by: Sean Paul <sean@poorly.run>
Link: https://patchwork.freedesktop.org/patch/msgid/20191101153754.22803-1-robh@kernel.org
The "read-modify-write register index" function is declared with a
confusing prototype: the "mask" and "reg" arguments are swapped.
Fortunately, this does not affect callers so far. Both arguments are
u32, and the wrapper macros (ocelot_rmw_ix etc) have the arguments in
the correct order (the one from ocelot_io.c).
Signed-off-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Vladimir Oltean says:
====================
Bonding fixes for Ocelot switch
This series fixes 2 issues with bonding in a system that integrates the
ocelot driver, but the ports that are bonded do not actually belong to
ocelot.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
lag_upper_info may be NULL on slave removal.
Fixes: dc96ee3730 ("net: mscc: ocelot: add bonding support")
Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Signed-off-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The check that the event is actually for this device should be moved
from the "port" handler to the net device handler.
Otherwise the port handler will deny bonding configuration for other
net devices in the same system (like enetc in the LS1028A) that don't
have the lag_upper_info->tx_type restriction that ocelot has.
Fixes: dc96ee3730 ("net: mscc: ocelot: add bonding support")
Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Signed-off-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For vlan push action, if eswitch flow source capability is enabled, flow
source value compared with MLX5_VPORT_UPLINK enum, to determine uplink
port. This lead to syndrome in dmesg if try to add vlan push action.
For example:
$ tc filter add dev vxlan0 ingress protocol ip prio 1 flower \
enc_dst_port 4789 \
action tunnel_key unset pipe \
action vlan push id 20 pipe \
action mirred egress redirect dev ens1f0_0
$ dmesg
...
[ 2456.883693] mlx5_core 0000:82:00.0: mlx5_cmd_check:756:(pid 5273): SET_FLOW_TABLE_ENTRY(0x936) op_mod(0x0) failed, status bad parameter(0x3), syndrome (0xa9c090)
Use the correct enum value MLX5_FLOW_CONTEXT_FLOW_SOURCE_UPLINK.
Fixes: bb204dcf39fe ("net/mlx5e: Determine source port properly for vlan push action")
Signed-off-by: Dmytro Linkin <dmitrolin@mellanox.com>
Reviewed-by: Vlad Buslov <vladbu@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
The rewrite data was no freed.
Fixes: 9db810ed2d ("net/mlx5: DR, Expose steering action functionality")
Signed-off-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
The value is already the calculation so remove the log prefix.
Fixes: e52c280240 ("net/mlx5: E-Switch, Add chains and priorities")
Signed-off-by: Roi Dayan <roid@mellanox.com>
Reviewed-by: Eli Britstein <elibr@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
The HID descriptors for most Wacom devices oddly declare the serial
number and other related fields as signed integers. When these numbers
are ingested by the HID subsystem, they are automatically sign-extended
into 32-bit integers. We treat the fields as unsigned elsewhere in the
kernel and userspace, however, so this sign-extension causes problems.
In particular, the sign-extended tool ID sent to userspace as ABS_MISC
does not properly match unsigned IDs used by xf86-input-wacom and libwacom.
We introduce a function 'wacom_s32tou' that can undo the automatic sign
extension performed by 'hid_snto32'. We call this function when processing
the serial number and related fields to ensure that we are dealing with
and reporting the unsigned form. We opt to use this method rather than
adding a descriptor fixup in 'wacom_hid_usage_quirk' since it should be
more robust in the face of future devices.
Ref: https://github.com/linuxwacom/input-wacom/issues/134
Fixes: f85c9dc678 ("HID: wacom: generic: Support tool ID and additional tool types")
CC: <stable@vger.kernel.org> # v4.10+
Signed-off-by: Jason Gerecke <jason.gerecke@wacom.com>
Reviewed-by: Aaron Armstrong Skomra <aaron.skomra@wacom.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
This reverts commit 385857adb8.
Reason for revert: Root cause of this issue is found. The workaround is not needed anymore.
Signed-off-by: Zhan Liu <zhan.liu@amd.com>
Reviewed-by: Hersen Wu <hersenxs.wu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
Navi10 has 6 PHY, but Navi14 only has 5 PHY, that is
because there is no ENGINE_ID_DIGD in Navi14. Without
this patch, many HDMI related issues (e.g. HDMI S3
resume failure, HDMI pink screen on boot) will be
observed.
[How]
If "eng_id" is larger than ENGINE_ID_DIGD, then
add "eng_id" by 1.
Signed-off-by: Zhan Liu <zhan.liu@amd.com>
Reviewed-by: Hersen Wu <hersenxs.wu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
doing kthread_park()/unpark() from drm_sched_entity_fini
while GPU reset is in progress defeats all the purpose of
drm_sched_stop->kthread_park.
If drm_sched_entity_fini->kthread_unpark() happens AFTER
drm_sched_stop->kthread_park nothing prevents from another
(third) thread to keep submitting job to HW which will be
picked up by the unparked scheduler thread and try to submit
to HW but fail because the HW ring is deactivated.
[How]
grab the reset lock before calling drm_sched_entity_fini()
Signed-off-by: Shirish S <shirish.s@amd.com>
Suggested-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
These were not aligned for optimal performance for GPUVM.
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Merge more fixes from Andrew Morton:
"17 fixes"
Mostly mm fixes and one ocfs2 locking fix.
* emailed patches from Andrew Morton <akpm@linux-foundation.org>:
mm: memcontrol: fix network errors from failing __GFP_ATOMIC charges
mm/memory_hotplug: fix updating the node span
scripts/gdb: fix debugging modules compiled with hot/cold partitioning
mm: slab: make page_cgroup_ino() to recognize non-compound slab pages properly
MAINTAINERS: update information for "MEMORY MANAGEMENT"
dump_stack: avoid the livelock of the dump_lock
zswap: add Vitaly to the maintainers list
mm/page_alloc.c: ratelimit allocation failure warnings more aggressively
mm/khugepaged: fix might_sleep() warn with CONFIG_HIGHPTE=y
mm, vmstat: reduce zone->lock holding time by /proc/pagetypeinfo
mm, vmstat: hide /proc/pagetypeinfo from normal users
mm/mmu_notifiers: use the right return code for WARN_ON
ocfs2: protect extent tree in ocfs2_prepare_inode_for_write()
mm: thp: handle page cache THP correctly in PageTransCompoundMap
mm, meminit: recalculate pcpu batch and high limits after init completes
mm/gup_benchmark: fix MAP_HUGETLB case
mm: memcontrol: fix NULL-ptr deref in percpu stats flush
Following commit 73e86cb03c ("arm64: Move PTE_RDONLY bit handling out
of set_pte_at()"), the PTE_RDONLY bit is no longer managed by
set_pte_at() but built into the PAGE_* attribute definitions.
Consequently, pte_same() must include this bit when checking two PTEs
for equality.
Remove the arm64-specific pte_same() function, practically reverting
commit 747a70e60b ("arm64: Fix copy-on-write referencing in HugeTLB")
Fixes: 73e86cb03c ("arm64: Move PTE_RDONLY bit handling out of set_pte_at()")
Cc: <stable@vger.kernel.org> # 4.14.x-
Cc: Will Deacon <will@kernel.org>
Cc: Steve Capper <steve.capper@arm.com>
Reported-by: John Stultz <john.stultz@linaro.org>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will@kernel.org>
Doug Berger says:
====================
net: bcmgenet: restore internal EPHY support (part 2)
This is a follow up to my previous submission (see [1]).
The first commit provides what is intended to be a complete solution
for the issues that can result from insufficient clocking of the MAC
during reset of its state machines. It should be backported to the
stable releases.
It is intended to replace the partial solution of commit 1f51548627
("net: bcmgenet: soft reset 40nm EPHYs before MAC init") which is
reverted by the second commit of this series and should not be back-
ported as noted in [2].
The third commit corrects a timing hazard with a polled PHY that can
occur when the MAC resumes and also when a v3 internal EPHY is reset
by the change in commit 25382b991d ("net: bcmgenet: reset 40nm EPHY
on energy detect"). It is expected that commit 25382b991d be back-
ported to stable first before backporting this commit.
[1] https://lkml.org/lkml/2019/10/16/1706
[2] https://lkml.org/lkml/2019/10/31/749
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
The phy_init_hw() function may reset the PHY to a configuration
that does not match manual network settings stored in the phydev
structure. If the phy state machine is polled rather than event
driven this can create a timing hazard where the phy state machine
might alter the settings stored in the phydev structure from the
value read from the BMCR.
This commit follows invocations of phy_init_hw() by the bcmgenet
driver with invocations of the genphy_config_aneg() function to
ensure that the BMCR is written to match the settings held in the
phydev structure. This prevents the risk of manual settings being
accidentally altered.
Fixes: 1c1008c793 ("net: bcmgenet: add main driver file")
Signed-off-by: Doug Berger <opendmb@gmail.com>
Acked-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This reverts commit 1f51548627.
This commit improved the chances of the umac resetting cleanly by
ensuring that the PHY was restored to its normal operation prior
to resetting the umac. However, there were still cases when the
PHY might not be driving a Tx clock to the umac during this window
(e.g. when the PHY detects no link).
The previous commit now ensures that the unimac receives clocks
from the MAC during its reset window so this commit is no longer
needed. This commit also has an unintended negative impact on the
MDIO performance of the UniMAC MDIO interface because it is used
before the MDIO interrupts are reenabled, so it should be removed.
Signed-off-by: Doug Berger <opendmb@gmail.com>
Acked-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As noted in commit 28c2d1a7a0 ("net: bcmgenet: enable loopback
during UniMAC sw_reset") the UniMAC must be clocked while sw_reset
is asserted for its state machines to reset cleanly.
The transmit and receive clocks used by the UniMAC are derived from
the signals used on its PHY interface. The bcmgenet MAC can be
configured to work with different PHY interfaces including MII,
GMII, RGMII, and Reverse MII on internal and external interfaces.
Unfortunately for the UniMAC, when configured for MII the Tx clock
is always driven from the PHY which places it outside of the direct
control of the MAC.
The earlier commit enabled a local loopback mode within the UniMAC
so that the receive clock would be derived from the transmit clock
which addressed the observed issue with an external GPHY disabling
it's Rx clock. However, when a Tx clock is not available this
loopback is insufficient.
This commit implements a workaround that leverages the fact that
the MAC can reliably generate all of its necessary clocking by
enterring the external GPHY RGMII interface mode with the UniMAC in
local loopback during the sw_reset interval. Unfortunately, this
has the undesirable side efect of the RGMII GTXCLK signal being
driven during the same window.
In most configurations this is a benign side effect as the signal
is either not routed to a pin or is already expected to drive the
pin. The one exception is when an external MII PHY is expected to
drive the same pin with its TX_CLK output creating output driver
contention.
This commit exploits the IEEE 802.3 clause 22 standard defined
isolate mode to force an external MII PHY to present a high
impedance on its TX_CLK output during the window to prevent any
contention at the pin.
The MII interface is used internally with the 40nm internal EPHY
which agressively disables its clocks for power savings leading to
incomplete resets of the UniMAC and many instabilities observed
over the years. The workaround of this commit is expected to put
an end to those problems.
Fixes: 1c1008c793 ("net: bcmgenet: add main driver file")
Signed-off-by: Doug Berger <opendmb@gmail.com>
Acked-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
perf report/top:
Jiri Olsa:
- Fix time sorting for big numbers, i.e.:
perf report -s time -F time,overhead --stdio
was failing because the sort comparision routine was returning 'int' while
that particular -s key was int64_t, fix it.
perf scripting engines:
Steven Rostedt (VMware):
- Iterate on tep event arrays directly, fixing a bug when generating python/perl
source code from a perf.data file with more than one tracepoint event.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
-----BEGIN PGP SIGNATURE-----
iHUEABYIAB0WIQR2GiIUctdOfX2qHhGyPKLppCJ+JwUCXcGdpgAKCRCyPKLppCJ+
JwoXAQDeac29fMxv3fHcmkxjQnt5M0f3GV14+bT8qmUZhDgi0QD8DzVpa1iKF+cC
N26yDNU0zcj6Rs6WUWlBgyXfvrS4JQs=
=fbXe
-----END PGP SIGNATURE-----
Merge tag 'perf-urgent-for-mingo-5.4-20191105' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/urgent
Pull perf fixes from Arnaldo Carvalho de Melo:
perf report/top:
Jiri Olsa:
- Fix time sorting for big numbers, i.e.:
perf report -s time -F time,overhead --stdio
was failing because the sort comparision routine was returning 'int' while
that particular -s key was int64_t, fix it.
perf scripting engines:
Steven Rostedt (VMware):
- Iterate on tep event arrays directly, fixing a bug when generating python/perl
source code from a perf.data file with more than one tracepoint event.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
drm_self_refresh_helper_update_avg_times() was incorrectly accessing the
new incoming state after drm_atomic_helper_commit_hw_done(). But this
state might have already been superceeded by an !nonblock atomic update
resulting in dereferencing an already free'd crtc_state.
TODO I *think* this will more or less do the right thing.. althought I'm
not 100% sure if, for example, we enter psr in a nonblock commit, and
then leave psr in a !nonblock commit that overtakes the completion of
the nonblock commit. Not sure if this sort of scenario can happen in
practice. But not crashing is better than crashing, so I guess we
should either take this patch or rever the self-refresh helpers until
Sean can figure out a better solution.
Fixes: d4da4e3334 ("drm: Measure Self Refresh Entry/Exit times to avoid thrashing")
Cc: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Rob Clark <robdclark@chromium.org>
[seanpaul fixed up some checkpatch warns]
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20191104173737.142558-1-robdclark@gmail.com
The clean up commit 41672c0c24 ("ALSA: timer: Simplify error path in
snd_timer_open()") unified the error handling code paths with the
standard goto, but it introduced a subtle bug: the timer instance is
stored in snd_timer_open() incorrectly even if it returns an error.
This may eventually lead to UAF, as spotted by fuzzer.
The culprit is the snd_timer_open() code checks the
SNDRV_TIMER_IFLG_EXCLUSIVE flag with the common variable timeri.
This variable is supposed to be the newly created instance, but we
(ab-)used it for a temporary check before the actual creation of a
timer instance. After that point, there is another check for the max
number of instances, and it bails out if over the threshold. Before
the refactoring above, it worked fine because the code returned
directly from that point. After the refactoring, however, it jumps to
the unified error path that stores the timeri variable in return --
even if it returns an error. Unfortunately this stored value is kept
in the caller side (snd_timer_user_tselect()) in tu->timeri. This
causes inconsistency later, as if the timer was successfully
assigned.
In this patch, we fix it by not re-using timeri variable but a
temporary variable for testing the exclusive connection, so timeri
remains NULL at that point.
Fixes: 41672c0c24 ("ALSA: timer: Simplify error path in snd_timer_open()")
Reported-and-tested-by: Tristan Madani <tristmd@gmail.com>
Cc: <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/20191106165547.23518-1-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
While upgrading from 4.16 to 5.2, we noticed these allocation errors in
the log of the new kernel:
SLUB: Unable to allocate memory on node -1, gfp=0xa20(GFP_ATOMIC)
cache: tw_sock_TCPv6(960:helper-logs), object size: 232, buffer size: 240, default order: 1, min order: 0
node 0: slabs: 5, objs: 170, free: 0
slab_out_of_memory+1
___slab_alloc+969
__slab_alloc+14
kmem_cache_alloc+346
inet_twsk_alloc+60
tcp_time_wait+46
tcp_fin+206
tcp_data_queue+2034
tcp_rcv_state_process+784
tcp_v6_do_rcv+405
__release_sock+118
tcp_close+385
inet_release+46
__sock_release+55
sock_close+17
__fput+170
task_work_run+127
exit_to_usermode_loop+191
do_syscall_64+212
entry_SYSCALL_64_after_hwframe+68
accompanied by an increase in machines going completely radio silent
under memory pressure.
One thing that changed since 4.16 is e699e2c6a6 ("net, mm: account
sock objects to kmemcg"), which made these slab caches subject to cgroup
memory accounting and control.
The problem with that is that cgroups, unlike the page allocator, do not
maintain dedicated atomic reserves. As a cgroup's usage hovers at its
limit, atomic allocations - such as done during network rx - can fail
consistently for extended periods of time. The kernel is not able to
operate under these conditions.
We don't want to revert the culprit patch, because it indeed tracks a
potentially substantial amount of memory used by a cgroup.
We also don't want to implement dedicated atomic reserves for cgroups.
There is no point in keeping a fixed margin of unused bytes in the
cgroup's memory budget to accomodate a consumer that is impossible to
predict - we'd be wasting memory and get into configuration headaches,
not unlike what we have going with min_free_kbytes. We do this for
physical mem because we have to, but cgroups are an accounting game.
Instead, account these privileged allocations to the cgroup, but let
them bypass the configured limit if they have to. This way, we get the
benefits of accounting the consumed memory and have it exert pressure on
the rest of the cgroup, but like with the page allocator, we shift the
burden of reclaimining on behalf of atomic allocations onto the regular
allocations that can block.
Link: http://lkml.kernel.org/r/20191022233708.365764-1-hannes@cmpxchg.org
Fixes: e699e2c6a6 ("net, mm: account sock objects to kmemcg")
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Suleiman Souhlal <suleiman@google.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: <stable@vger.kernel.org> [4.18+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We recently started updating the node span based on the zone span to
avoid touching uninitialized memmaps.
Currently, we will always detect the node span to start at 0, meaning a
node can easily span too many pages. pgdat_is_empty() will still work
correctly if all zones span no pages. We should skip over all zones
without spanned pages and properly handle the first detected zone that
spans pages.
Unfortunately, in contrast to the zone span (/proc/zoneinfo), the node
span cannot easily be inspected and tested. The node span gives no real
guarantees when an architecture supports memory hotplug, meaning it can
easily contain holes or span pages of different nodes.
The node span is not really used after init on architectures that
support memory hotplug.
E.g., we use it in mm/memory_hotplug.c:try_offline_node() and in
mm/kmemleak.c:kmemleak_scan(). These users seem to be fine.
Link: http://lkml.kernel.org/r/20191027222714.5313-1-david@redhat.com
Fixes: 00d6c019b5 ("mm/memory_hotplug: don't access uninitialized memmaps in shrink_pgdat_span()")
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
gcc's -freorder-blocks-and-partition option makes it group frequently
and infrequently used code in .text.hot and .text.unlikely sections
respectively. At least when building modules on s390, this option is
used by default.
gdb assumes that all code is located in .text section, and that .text
section is located at module load address. With such modules this is no
longer the case: there is code in .text.hot and .text.unlikely, and
either of them might precede .text.
Fix by explicitly telling gdb the addresses of code sections.
It might be tempting to do this for all sections, not only the ones in
the white list. Unfortunately, gdb appears to have an issue, when
telling it about e.g. loadable .note.gnu.build-id section causes it to
think that non-loadable .note.Linux section is loaded at address 0,
which in turn causes NULL pointers to be resolved to bogus symbols. So
keep using the white list approach for the time being.
Link: http://lkml.kernel.org/r/20191028152734.13065-1-iii@linux.ibm.com
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Reviewed-by: Jan Kiszka <jan.kiszka@siemens.com>
Cc: Kieran Bingham <kbingham@kernel.org>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
page_cgroup_ino() doesn't return a valid memcg pointer for non-compound
slab pages, because it depends on PgHead AND PgSlab flags to be set to
determine the memory cgroup from the kmem_cache. It's correct for
compound pages, but not for generic small pages. Those don't have PgHead
set, so it ends up returning zero.
Fix this by replacing the condition to PageSlab() && !PageTail().
Before this patch:
[root@localhost ~]# ./page-types -c /sys/fs/cgroup/user.slice/user-0.slice/user@0.service/ | grep slab
0x0000000000000080 38 0 _______S___________________________________ slab
After this patch:
[root@localhost ~]# ./page-types -c /sys/fs/cgroup/user.slice/user-0.slice/user@0.service/ | grep slab
0x0000000000000080 147 0 _______S___________________________________ slab
Also, hwpoison_filter_task() uses output of page_cgroup_ino() in order
to filter error injection events based on memcg. So if
page_cgroup_ino() fails to return memcg pointer, we just fail to inject
memory error. Considering that hwpoison filter is for testing, affected
users are limited and the impact should be marginal.
[n-horiguchi@ah.jp.nec.com: changelog additions]
Link: http://lkml.kernel.org/r/20191031012151.2722280-1-guro@fb.com
Fixes: 4d96ba3530 ("mm: memcg/slab: stop setting page->mem_cgroup pointer for slab pages")
Signed-off-by: Roman Gushchin <guro@fb.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Daniel Jordan <daniel.m.jordan@oracle.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
I was trying to find the mm tree in MAINTAINERS by searching "Morton".
Unfortunately, I didn't find one. And I didn't even locate the MEMORY
MANAGEMENT section quickly, because Andrew's name was not listed there.
Thanks to Johannes who helped me find the mm tree.
Let save other's time searching around by adding:
M: Andrew Morton <akpm@linux-foundation.org>
T: git git://github.com/hnaz/linux-mm.git
[akpm@linux-foundation.org: add ozlabs.org quilt trees]
Link: http://lkml.kernel.org/r/20191030202217.3498133-1-songliubraving@fb.com
Signed-off-by: Song Liu <songliubraving@fb.com>
Acked-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In the current code, we use the atomic_cmpxchg() to serialize the output
of the dump_stack(), but this implementation suffers the thundering herd
problem. We have observed such kind of livelock on a Marvell cn96xx
board(24 cpus) when heavily using the dump_stack() in a kprobe handler.
Actually we can let the competitors to wait for the releasing of the
lock before jumping to atomic_cmpxchg(). This will definitely mitigate
the thundering herd problem. Thanks Linus for the suggestion.
[akpm@linux-foundation.org: fix comment]
Link: http://lkml.kernel.org/r/20191030031637.6025-1-haokexin@gmail.com
Fixes: b58d977432 ("dump_stack: serialize the output from dump_stack()")
Signed-off-by: Kevin Hao <haokexin@gmail.com>
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Per conversation with Dan, add myself to the zswap MAINTAINERS list.
Link: http://lkml.kernel.org/r/20191028143154.31304-1-vitaly.wool@konsulko.com
Signed-off-by: Vitaly Wool <vitaly.wool@konsulko.com>
Acked-by: Dan Streetman <ddstreet@ieee.org>
Acked-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
While investigating a bug related to higher atomic allocation failures,
we noticed the failure warnings positively drowning the console, and in
our case trigger lockup warnings because of a serial console too slow to
handle all that output.
But even if we had a faster console, it's unclear what additional
information the current level of repetition provides.
Allocation failures happen for three reasons: The machine is OOM, the VM
is failing to handle reasonable requests, or somebody is making
unreasonable requests (and didn't acknowledge their opportunism with
__GFP_NOWARN). Having the memory dump, a callstack, and the ratelimit
stats on skipped failure warnings should provide enough information to
let users/admins/developers know whether something is wrong and point
them in the right direction for debugging, bpftracing etc.
Limit allocation failure warnings to one spew every ten seconds.
Link: http://lkml.kernel.org/r/20191028194906.26899-1-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
I got some khugepaged spew on a 32bit x86:
BUG: sleeping function called from invalid context at include/linux/mmu_notifier.h:346
in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 25, name: khugepaged
INFO: lockdep is turned off.
CPU: 1 PID: 25 Comm: khugepaged Not tainted 5.4.0-rc5-elk+ #206
Hardware name: System manufacturer P5Q-EM/P5Q-EM, BIOS 2203 07/08/2009
Call Trace:
dump_stack+0x66/0x8e
___might_sleep.cold.96+0x95/0xa6
__might_sleep+0x2e/0x80
collapse_huge_page.isra.51+0x5ac/0x1360
khugepaged+0x9a9/0x20f0
kthread+0xf5/0x110
ret_from_fork+0x2e/0x38
Looks like it's due to CONFIG_HIGHPTE=y pte_offset_map()->kmap_atomic()
vs. mmu_notifier_invalidate_range_start(). Let's do the naive approach
and just reorder the two operations.
Link: http://lkml.kernel.org/r/20191029201513.GG1208@intel.com
Fixes: 810e24e009 ("mm/mmu_notifiers: annotate with might_sleep()")
Signed-off-by: Ville Syrjl <ville.syrjala@linux.intel.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Ira Weiny <ira.weiny@intel.com>
Cc: Jason Gunthorpe <jgg@mellanox.com>
Cc: Daniel Vetter <daniel.vetter@intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
pagetypeinfo_showfree_print is called by zone->lock held in irq mode.
This is not really nice because it blocks both any interrupts on that
cpu and the page allocator. On large machines this might even trigger
the hard lockup detector.
Considering the pagetypeinfo is a debugging tool we do not really need
exact numbers here. The primary reason to look at the outuput is to see
how pageblocks are spread among different migratetypes and low number of
pages is much more interesting therefore putting a bound on the number
of pages on the free_list sounds like a reasonable tradeoff.
The new output will simply tell
[...]
Node 6, zone Normal, type Movable >100000 >100000 >100000 >100000 41019 31560 23996 10054 3229 983 648
instead of
Node 6, zone Normal, type Movable 399568 294127 221558 102119 41019 31560 23996 10054 3229 983 648
The limit has been chosen arbitrary and it is a subject of a future
change should there be a need for that.
While we are at it, also drop the zone lock after each free_list
iteration which will help with the IRQ and page allocator responsiveness
even further as the IRQ lock held time is always bound to those 100k
pages.
[akpm@linux-foundation.org: tweak comment text, per David Hildenbrand]
Link: http://lkml.kernel.org/r/20191025072610.18526-3-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Suggested-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Waiman Long <longman@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: David Hildenbrand <david@redhat.com>
Acked-by: Rafael Aquini <aquini@redhat.com>
Acked-by: David Rientjes <rientjes@google.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Jann Horn <jannh@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Roman Gushchin <guro@fb.com>
Cc: Song Liu <songliubraving@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
/proc/pagetypeinfo is a debugging tool to examine internal page
allocator state wrt to fragmentation. It is not very useful for any
other use so normal users really do not need to read this file.
Waiman Long has noticed that reading this file can have negative side
effects because zone->lock is necessary for gathering data and that a)
interferes with the page allocator and its users and b) can lead to hard
lockups on large machines which have very long free_list.
Reduce both issues by simply not exporting the file to regular users.
Link: http://lkml.kernel.org/r/20191025072610.18526-2-mhocko@kernel.org
Fixes: 467c996c1e ("Print out statistics in relation to fragmentation avoidance to /proc/pagetypeinfo")
Signed-off-by: Michal Hocko <mhocko@suse.com>
Reported-by: Waiman Long <longman@redhat.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Waiman Long <longman@redhat.com>
Acked-by: Rafael Aquini <aquini@redhat.com>
Acked-by: David Rientjes <rientjes@google.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Roman Gushchin <guro@fb.com>
Cc: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Cc: Jann Horn <jannh@google.com>
Cc: Song Liu <songliubraving@fb.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The return code from the op callback is actually in _ret, while the
WARN_ON was checking ret which causes it to misfire.
Link: http://lkml.kernel.org/r/20191025175502.GA31127@ziepe.ca
Fixes: 8402ce61be ("mm/mmu_notifiers: check if mmu notifier callbacks are allowed to fail")
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When the extent tree is modified, it should be protected by inode
cluster lock and ip_alloc_sem.
The extent tree is accessed and modified in the
ocfs2_prepare_inode_for_write, but isn't protected by ip_alloc_sem.
The following is a case. The function ocfs2_fiemap is accessing the
extent tree, which is modified at the same time.
kernel BUG at fs/ocfs2/extent_map.c:475!
invalid opcode: 0000 [#1] SMP
Modules linked in: tun ocfs2 ocfs2_nodemanager configfs ocfs2_stackglue [...]
CPU: 16 PID: 14047 Comm: o2info Not tainted 4.1.12-124.23.1.el6uek.x86_64 #2
Hardware name: Oracle Corporation ORACLE SERVER X7-2L/ASM, MB MECH, X7-2L, BIOS 42040600 10/19/2018
task: ffff88019487e200 ti: ffff88003daa4000 task.ti: ffff88003daa4000
RIP: ocfs2_get_clusters_nocache.isra.11+0x390/0x550 [ocfs2]
Call Trace:
ocfs2_fiemap+0x1e3/0x430 [ocfs2]
do_vfs_ioctl+0x155/0x510
SyS_ioctl+0x81/0xa0
system_call_fastpath+0x18/0xd8
Code: 18 48 c7 c6 60 7f 65 a0 31 c0 bb e2 ff ff ff 48 8b 4a 40 48 8b 7a 28 48 c7 c2 78 2d 66 a0 e8 38 4f 05 00 e9 28 fe ff ff 0f 1f 00 <0f> 0b 66 0f 1f 44 00 00 bb 86 ff ff ff e9 13 fe ff ff 66 0f 1f
RIP ocfs2_get_clusters_nocache.isra.11+0x390/0x550 [ocfs2]
---[ end trace c8aa0c8180e869dc ]---
Kernel panic - not syncing: Fatal exception
Kernel Offset: disabled
This issue can be reproduced every week in a production environment.
This issue is related to the usage mode. If others use ocfs2 in this
mode, the kernel will panic frequently.
[akpm@linux-foundation.org: coding style fixes]
[Fix new warning due to unused function by removing said function - Linus ]
Link: http://lkml.kernel.org/r/1568772175-2906-2-git-send-email-sunny.s.zhang@oracle.com
Signed-off-by: Shuning Zhang <sunny.s.zhang@oracle.com>
Reviewed-by: Junxiao Bi <junxiao.bi@oracle.com>
Reviewed-by: Gang He <ghe@suse.com>
Cc: Mark Fasheh <mark@fasheh.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Joseph Qi <jiangqi903@gmail.com>
Cc: Changwei Ge <gechangwei@live.cn>
Cc: Jun Piao <piaojun@huawei.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We have a usecase to use tmpfs as QEMU memory backend and we would like
to take the advantage of THP as well. But, our test shows the EPT is
not PMD mapped even though the underlying THP are PMD mapped on host.
The number showed by /sys/kernel/debug/kvm/largepage is much less than
the number of PMD mapped shmem pages as the below:
7f2778200000-7f2878200000 rw-s 00000000 00:14 262232 /dev/shm/qemu_back_mem.mem.Hz2hSf (deleted)
Size: 4194304 kB
[snip]
AnonHugePages: 0 kB
ShmemPmdMapped: 579584 kB
[snip]
Locked: 0 kB
cat /sys/kernel/debug/kvm/largepages
12
And some benchmarks do worse than with anonymous THPs.
By digging into the code we figured out that commit 127393fbe5 ("mm:
thp: kvm: fix memory corruption in KVM with THP enabled") checks if
there is a single PTE mapping on the page for anonymous THP when setting
up EPT map. But the _mapcount < 0 check doesn't work for page cache THP
since every subpage of page cache THP would get _mapcount inc'ed once it
is PMD mapped, so PageTransCompoundMap() always returns false for page
cache THP. This would prevent KVM from setting up PMD mapped EPT entry.
So we need handle page cache THP correctly. However, when page cache
THP's PMD gets split, kernel just remove the map instead of setting up
PTE map like what anonymous THP does. Before KVM calls get_user_pages()
the subpages may get PTE mapped even though it is still a THP since the
page cache THP may be mapped by other processes at the mean time.
Checking its _mapcount and whether the THP has PTE mapped or not.
Although this may report some false negative cases (PTE mapped by other
processes), it looks not trivial to make this accurate.
With this fix /sys/kernel/debug/kvm/largepage would show reasonable
pages are PMD mapped by EPT as the below:
7fbeaee00000-7fbfaee00000 rw-s 00000000 00:14 275464 /dev/shm/qemu_back_mem.mem.SKUvat (deleted)
Size: 4194304 kB
[snip]
AnonHugePages: 0 kB
ShmemPmdMapped: 557056 kB
[snip]
Locked: 0 kB
cat /sys/kernel/debug/kvm/largepages
271
And the benchmarks are as same as anonymous THPs.
[yang.shi@linux.alibaba.com: v4]
Link: http://lkml.kernel.org/r/1571865575-42913-1-git-send-email-yang.shi@linux.alibaba.com
Link: http://lkml.kernel.org/r/1571769577-89735-1-git-send-email-yang.shi@linux.alibaba.com
Fixes: dd78fedde4 ("rmap: support file thp")
Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
Reported-by: Gang Deng <gavin.dg@linux.alibaba.com>
Tested-by: Gang Deng <gavin.dg@linux.alibaba.com>
Suggested-by: Hugh Dickins <hughd@google.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: <stable@vger.kernel.org> [4.8+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Deferred memory initialisation updates zone->managed_pages during the
initialisation phase but before that finishes, the per-cpu page
allocator (pcpu) calculates the number of pages allocated/freed in
batches as well as the maximum number of pages allowed on a per-cpu
list. As zone->managed_pages is not up to date yet, the pcpu
initialisation calculates inappropriately low batch and high values.
This increases zone lock contention quite severely in some cases with
the degree of severity depending on how many CPUs share a local zone and
the size of the zone. A private report indicated that kernel build
times were excessive with extremely high system CPU usage. A perf
profile indicated that a large chunk of time was lost on zone->lock
contention.
This patch recalculates the pcpu batch and high values after deferred
initialisation completes for every populated zone in the system. It was
tested on a 2-socket AMD EPYC 2 machine using a kernel compilation
workload -- allmodconfig and all available CPUs.
mmtests configuration: config-workload-kernbench-max Configuration was
modified to build on a fresh XFS partition.
kernbench
5.4.0-rc3 5.4.0-rc3
vanilla resetpcpu-v2
Amean user-256 13249.50 ( 0.00%) 16401.31 * -23.79%*
Amean syst-256 14760.30 ( 0.00%) 4448.39 * 69.86%*
Amean elsp-256 162.42 ( 0.00%) 119.13 * 26.65%*
Stddev user-256 42.97 ( 0.00%) 19.15 ( 55.43%)
Stddev syst-256 336.87 ( 0.00%) 6.71 ( 98.01%)
Stddev elsp-256 2.46 ( 0.00%) 0.39 ( 84.03%)
5.4.0-rc3 5.4.0-rc3
vanilla resetpcpu-v2
Duration User 39766.24 49221.79
Duration System 44298.10 13361.67
Duration Elapsed 519.11 388.87
The patch reduces system CPU usage by 69.86% and total build time by
26.65%. The variance of system CPU usage is also much reduced.
Before, this was the breakdown of batch and high values over all zones
was:
256 batch: 1
256 batch: 63
512 batch: 7
256 high: 0
256 high: 378
512 high: 42
512 pcpu pagesets had a batch limit of 7 and a high limit of 42. After
the patch:
256 batch: 1
768 batch: 63
256 high: 0
768 high: 378
[mgorman@techsingularity.net: fix merge/linkage snafu]
Link: http://lkml.kernel.org/r/20191023084705.GD3016@techsingularity.netLink: http://lkml.kernel.org/r/20191021094808.28824-2-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Matt Fleming <matt@codeblueprint.co.uk>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Qian Cai <cai@lca.pw>
Cc: <stable@vger.kernel.org> [4.1+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
