The --for-each-cgroup can have the same cgroup multiple times, but this
confuses BPF counters (since they have the same cgroup id), making only
the last cgroup events to be counted.
Let's check the cgroup name before adding a new entry to the cgroups
list.
Before:
$ sudo ./perf stat -a --bpf-counters --for-each-cgroup /,/ sleep 1
Performance counter stats for 'system wide':
<not counted> msec cpu-clock /
<not counted> context-switches /
<not counted> cpu-migrations /
<not counted> page-faults /
<not counted> cycles /
<not counted> instructions /
<not counted> branches /
<not counted> branch-misses /
8,016.04 msec cpu-clock / # 7.998 CPUs utilized
6,152 context-switches / # 767.461 /sec
250 cpu-migrations / # 31.187 /sec
442 page-faults / # 55.139 /sec
613,111,487 cycles / # 0.076 GHz
280,599,604 instructions / # 0.46 insn per cycle
57,692,724 branches / # 7.197 M/sec
3,385,168 branch-misses / # 5.87% of all branches
1.002220125 seconds time elapsed
After it becomes similar to the non-BPF mode:
$ sudo ./perf stat -a --bpf-counters --for-each-cgroup /,/ sleep 1
Performance counter stats for 'system wide':
8,013.38 msec cpu-clock / # 7.998 CPUs utilized
6,859 context-switches / # 855.944 /sec
334 cpu-migrations / # 41.680 /sec
345 page-faults / # 43.053 /sec
782,326,119 cycles / # 0.098 GHz
471,645,724 instructions / # 0.60 insn per cycle
94,963,430 branches / # 11.851 M/sec
3,685,511 branch-misses / # 3.88% of all branches
1.001864539 seconds time elapsed
Committer notes:
As a reminder, to test with BPF counters one has to use BUILD_BPF_SKEL=1
in the make command line and have clang/llvm installed when building
perf, otherwise the --bpf-counters option will not be available:
# perf stat -a --bpf-counters --for-each-cgroup /,/ sleep 1
Error: unknown option `bpf-counters'
Usage: perf stat [<options>] [<command>]
-a, --all-cpus system-wide collection from all CPUs
<SNIP>
#
Fixes: bb1c15b60b ("perf stat: Support regex pattern in --for-each-cgroup")
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: bpf@vger.kernel.org
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/r/20230104064402.1551516-5-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
UAPI Changes:
* connector: Support analog-TV mode property
* media: Add MEDIA_BUS_FMT_RGB565_1X24_CPADHI,
MEDIA_BUS_FMT_RGB666_1X18 and MEDIA_BUS_FMT_RGB666_1X24_CPADHI
Cross-subsystem Changes:
* dma-buf: Documentation fixes
* i2c: Introduce i2c_client_get_device_id() helper
Core Changes:
* Improve support for analog TV output
* bridge: Remove unused drm_bridge_chain functions
* debugfs: Add per-device helpers and convert various DRM drivers
* dp-mst: Various fixes
* fbdev emulation: Always pick 32 bpp as default
* KUnit: Add tests for managed helpers; Various cleanups
* panel-orientation: Add quirks for Lenovo Yoga Tab 3 X90F and DynaBook K50
* TTM: Open-code ttm_bo_wait() and remove the helper
Driver Changes:
* Fix preferred depth and bpp values throughout DRM drivers
* Remove #CONFIG_PM guards throughout DRM drivers
* ast: Various fixes
* bridge: Implement i2c's probe_new in various drivers; Fixes; ite-it6505:
Locking fixes, Cache EDID data; ite-it66121: Support IT6610 chip,
Cleanups; lontium-tl9611: Fix HDMI on DragonBoard 845c; parade-ps8640:
Use atomic bridge functions
* gud: Convert to DRM shadow-plane helpers; Perform flushing synchronously
during atomic update
* ili9486: Support 16-bit pixel data
* imx: Split off IPUv3 driver; Various fixes
* mipi-dbi: Convert to DRM shadow-plane helpers plus rsp driver changes;i
Support separate I/O-voltage supply
* mxsfb: Depend on ARCH_MXS or ARCH_MXC
* omapdrm: Various fixes
* panel: Use ktime_get_boottime() to measure power-down delay in various
drivers; Fix auto-suspend delay in various drivers; orisetech-ota5601a:
Add support
* sprd: Cleanups
* sun4i: Convert to new TV-mode property
* tidss: Various fixes
* v3d: Various fixes
* vc4: Convert to new TV-mode property; Support Kunit tests; Cleanups;
dpi: Support RGB565 and RGB666 formats; dsi: Convert DSI driver to
bridge
* virtio: Improve tracing
* vkms: Support small cursors in IGT tests; Various fixes
-----BEGIN PGP SIGNATURE-----
iQEzBAABCAAdFiEEchf7rIzpz2NEoWjlaA3BHVMLeiMFAmO0B0QACgkQaA3BHVML
eiMKtAf9EXt5yaEonR4gsGNVD70VkebW+jPEWbEMg5hMFDHE9sjBja8T7bOxeeL1
BVofJiAuEZW9176eqeWvShuwOuiE7lf4WQLMvXfFmtHF/Nac9HUtEvOcvc1vEDUB
y6VFlVHe8mSp+Iy0WPLyZCtT4d7v2eM+VYm0HgFa74dSTzQTLGPiUI/XVb/YDaA7
FHKB5NvsMX9S1XvjxCsq0jA5bo8SSzh5CVKerdAZkBJMCSmKiY09o2p5C7vw3EAz
WUXL5CXqd9bwgvZa/9ZClQtpJaikNGoOaMSEl67rbWdVTk0LIpqx3nTY15py5LXd
SG4Wtfor3Vf1REa9TrR2uCIOmh+1gQ==
=0/PC
-----END PGP SIGNATURE-----
Merge tag 'drm-misc-next-2023-01-03' of git://anongit.freedesktop.org/drm/drm-misc into drm-next
drm-misc-next for v6.3:
UAPI Changes:
* connector: Support analog-TV mode property
* media: Add MEDIA_BUS_FMT_RGB565_1X24_CPADHI,
MEDIA_BUS_FMT_RGB666_1X18 and MEDIA_BUS_FMT_RGB666_1X24_CPADHI
Cross-subsystem Changes:
* dma-buf: Documentation fixes
* i2c: Introduce i2c_client_get_device_id() helper
Core Changes:
* Improve support for analog TV output
* bridge: Remove unused drm_bridge_chain functions
* debugfs: Add per-device helpers and convert various DRM drivers
* dp-mst: Various fixes
* fbdev emulation: Always pick 32 bpp as default
* KUnit: Add tests for managed helpers; Various cleanups
* panel-orientation: Add quirks for Lenovo Yoga Tab 3 X90F and DynaBook K50
* TTM: Open-code ttm_bo_wait() and remove the helper
Driver Changes:
* Fix preferred depth and bpp values throughout DRM drivers
* Remove #CONFIG_PM guards throughout DRM drivers
* ast: Various fixes
* bridge: Implement i2c's probe_new in various drivers; Fixes; ite-it6505:
Locking fixes, Cache EDID data; ite-it66121: Support IT6610 chip,
Cleanups; lontium-tl9611: Fix HDMI on DragonBoard 845c; parade-ps8640:
Use atomic bridge functions
* gud: Convert to DRM shadow-plane helpers; Perform flushing synchronously
during atomic update
* ili9486: Support 16-bit pixel data
* imx: Split off IPUv3 driver; Various fixes
* mipi-dbi: Convert to DRM shadow-plane helpers plus rsp driver changes;i
Support separate I/O-voltage supply
* mxsfb: Depend on ARCH_MXS or ARCH_MXC
* omapdrm: Various fixes
* panel: Use ktime_get_boottime() to measure power-down delay in various
drivers; Fix auto-suspend delay in various drivers; orisetech-ota5601a:
Add support
* sprd: Cleanups
* sun4i: Convert to new TV-mode property
* tidss: Various fixes
* v3d: Various fixes
* vc4: Convert to new TV-mode property; Support Kunit tests; Cleanups;
dpi: Support RGB565 and RGB666 formats; dsi: Convert DSI driver to
bridge
* virtio: Improve tracing
* vkms: Support small cursors in IGT tests; Various fixes
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
From: Thomas Zimmermann <tzimmermann@suse.de>
Link: https://patchwork.freedesktop.org/patch/msgid/Y7QIwlfElAYWxRcR@linux-uq9g
When --for-each-cgroup option is used, it fails when any of events is
not supported and exits immediately. This is not how 'perf stat'
handles unsupported events.
Let's ignore the failure and proceed with others so that the output is
similar to when BPF counters are not used:
Before:
$ sudo ./perf stat -a --bpf-counters -e L1-icache-loads,L1-dcache-loads --for-each-cgroup system.slice,user.slice sleep 1
Failed to open first cgroup events
$
After it shows output similat to when --bpf-counters isn't specified:
$ sudo ./perf stat -a --bpf-counters -e L1-icache-loads,L1-dcache-loads --for-each-cgroup system.slice,user.slice sleep 1
Performance counter stats for 'system wide':
<not supported> L1-icache-loads system.slice
29,892,418 L1-dcache-loads system.slice
<not supported> L1-icache-loads user.slice
52,497,220 L1-dcache-loads user.slice
$
Fixes: 944138f048 ("perf stat: Enable BPF counter with --for-each-cgroup")
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/r/20230104064402.1551516-4-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
perf test '84: probe libc's inet_pton & backtrace it with ping' fails on
s390. Debugging revealed a changed stack trace for the ping command
using probes:
ping 35729 [002] 8006.365063: probe_libc:inet_pton: (3ff9603e7c0)
13e7c0 __GI___inet_pton+0x0 (/usr/lib64/libc.so.6)
---> 104371 text_to_binary_address+0xef1 (inlined)
104371 gaih_inet+0xef1 (inlined)
104371 __GI_getaddrinfo+0xef1 (inlined)
5d4b main+0x139b (/usr/bin/ping)
The line "---> text_to_binary_address ..." is new. It was introduced
with glibc version 2.36.7.2 released with Fedora 37 for s390.
Output before
# perf test inet_pton
84: probe libc's inet_pton & backtrace it with ping : FAILED!
#
Output after:
# perf test inet_pton
84: probe libc's inet_pton & backtrace it with ping : Ok
#
Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Sumanth Korikkar <sumanthk@linux.ibm.com>
Cc: Sven Schnelle <svens@linux.ibm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Link: https://lore.kernel.org/r/20221228145704.2702487-1-tmricht@linux.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The busy timeout logic checks for the AUX BUSY, then waits for the
timeout period and then after timeout reads the register for BUSY or
Success.
Instead replace interrupt with polling so as to read the AUX CTL
register often before the timeout period. Looks like there might be some
issue with interrupt-on-read. Hence changing the logic to polling read.
v2: replace interrupt with polling read
v3: use usleep_rang instead of msleep, updated commit msg
v4: use intel_wait_for_regiter internal function
v5: use __intel_de_wait_for_register with 500us slow and 10ms fast timeout
v6: check return value of __intel_de_wait_for_register
v7: using default 2us for intel_de_wait_for_register
Signed-off-by: Arun R Murthy <arun.r.murthy@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221221033209.1284435-1-arun.r.murthy@intel.com
a474d3fbe2 ("PCI/MSI: Get rid of PCI_MSI_IRQ_DOMAIN") removed
PCI_MSI_IRQ_DOMAIN and changed all references to refer to PCI_MSI instead.
ba6ed462dc ("PCI: dwc: Add Baikal-T1 PCIe controller support")
independently added PCIE_BT1, depending on PCI_MSI_IRQ_DOMAIN.
Both commits appeared in v6.2-rc1, so the latter missed the conversion from
PCI_MSI_IRQ_DOMAIN to PCI_MSI. Update PCIE_BT1 to depend on PCI_MSI
instead.
[bhelgaas: commit log]
Link: https://lore.kernel.org/r/20221215103452.23131-1-lukas.bulwahn@gmail.com
Signed-off-by: Lukas Bulwahn <lukas.bulwahn@gmail.com>
Signed-off-by: Lorenzo Pieralisi <lpieralisi@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Serge Semin <fancer.lancer@gmail.com>
Without this, the CPUs are left in a random pstate. Since we don't
support deep idle yet (which powers down the CPUs), this results in
significantly increased idle power consumption in suspend.
Fixes: 6286bbb405 ("cpufreq: apple-soc: Add new driver to control Apple SoC CPU P-states")
Signed-off-by: Hector Martin <marcan@marcan.st>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
memblock_phys_free() is the counterpart to memblock_phys_alloc.
Change memblock_alloc_xx() with memblock_phys_alloc_xx() to keep
consistency.
Signed-off-by: Miaoqian Lin <linmq006@gmail.com>
Link: https://lore.kernel.org/r/20221216100304.688209-1-linmq006@gmail.com
Signed-off-by: Mike Rapoport (IBM) <rppt@kernel.org>
Commit cf4694be2b ("tools: Add atomic_test_and_set_bit()") changed
tools/arch/x86/include/asm/atomic.h to include <asm/asm.h>, which causes
'make -C tools/testing/memblock' to fail with:
In file included from ../../include/asm/atomic.h:6,
from ../../include/linux/atomic.h:5,
from ./linux/mmzone.h:5,
from ../../include/linux/mm.h:5,
from ../../include/linux/pfn.h:5,
from ./linux/memory_hotplug.h:6,
from ./linux/init.h:7,
from ./linux/memblock.h:11,
from tests/common.h:8,
from tests/basic_api.h:5,
from main.c:2:
../../include/asm/../../arch/x86/include/asm/atomic.h:11:10: fatal error: asm/asm.h: No such file or directory
11 | #include <asm/asm.h>
| ^~~~~~~~~~~
Create a symlink to asm/asm.h in the same manner as the existing one to
asm/cmpxchg.h.
Signed-off-by: Aaron Thompson <dev@aaront.org>
Link: https://lore.kernel.org/r/010101857c402765-96e2dbc6-b82b-47e2-a437-4834dbe0b96b-000000@us-west-2.amazonses.com
Signed-off-by: Mike Rapoport (IBM) <rppt@kernel.org>
We missed the window between the TIF flag update and the next reschedule.
Signed-off-by: Rodrigo Branco <bsdaemon@google.com>
Reviewed-by: Borislav Petkov (AMD) <bp@alien8.de>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: <stable@vger.kernel.org>
Tony Nguyen says:
====================
Intel Wired LAN Driver Updates 2023-01-03 (igc)
Muhammad Husaini Zulkifli says:
Improvements to the Time-Sensitive Networking (TSN) Qbv Scheduling
capabilities were included in this patch series for I226 SKU.
An overview of each patch series is given below:
Patch 1: To enable basetime scheduling in the future, remove the existing
restriction for i226 stepping while maintain the restriction for i225.
Patch 2: Remove the restriction which require a controller reset when
setting the basetime register for new i226 steps and enable the second
GCL configuration.
Patch 3: Remove the power reset adapter during disabling the tsn config.
---
Patches remaining from initial PR:
https://lore.kernel.org/netdev/20221205212414.3197525-1-anthony.l.nguyen@intel.com/
after sending net patches:
https://lore.kernel.org/netdev/20221215230758.3595578-1-anthony.l.nguyen@intel.com/
Note: patch 3 is an additional patch from the initial PR.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Although we applied a workaround for the hw constraints code with the
implicit feedback sync, it still has a potential problem. Namely, as
the code treats only the first matching (sync) endpoint, it might be
too restrictive when multiple endpoints are listed in the substream's
format list.
This patch is another attempt to improve the hw constraint handling
for the implicit feedback sync. The code is rewritten and the sync EP
handling for the rate and the format is put inside the fmt_list loop
in each hw_rule_*() function instead of the additional rules. The
rules for the period size and periods are extended to loop over the
fmt_list like others, and they apply the constraints only if needed.
Link: https://lore.kernel.org/r/4e509aea-e563-e592-e652-ba44af6733fe@veniogames.com
Link: https://lore.kernel.org/r/20230102170759.29610-4-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The fix commit the commit e4ea77f8e5 ("ALSA: usb-audio: Always apply
the hw constraints for implicit fb sync") tried to address the bug
where an incorrect PCM parameter is chosen when two (implicit fb)
streams are set up at the same time. This change had, however, some
side effect: once when the sync endpoint is chosen and set up, this
restriction is applied at the next hw params unless it's freed via hw
free explicitly.
This patch is a workaround for the problem by relaxing the hw
constraints a bit for the implicit fb sync. We still keep applying
the hw constraints for implicit fb sync, but only when the matching
sync EP is being used by other streams.
Fixes: e4ea77f8e5 ("ALSA: usb-audio: Always apply the hw constraints for implicit fb sync")
Reported-by: Ruud van Asseldonk <ruud@veniogames.com>
Link: https://lore.kernel.org/r/4e509aea-e563-e592-e652-ba44af6733fe@veniogames.com
Link: https://lore.kernel.org/r/20230102170759.29610-3-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
At the PCM hw params, we may re-configure the endpoints and it's done
by a temporary EP close followed by re-open. A potential problem
there is that the EP might be already running internally at the PCM
prepare stage; it's seen typically in the playback stream with the
implicit feedback sync. As this stream start isn't tracked by the
core PCM layer, we'd need to stop it explicitly, and that's the
missing piece.
This patch adds the stop_endpoints() call at snd_usb_hw_params() to
assure the stream stop before closing the EPs.
Fixes: bf6313a0ff ("ALSA: usb-audio: Refactor endpoint management")
Link: https://lore.kernel.org/r/4e509aea-e563-e592-e652-ba44af6733fe@veniogames.com
Link: https://lore.kernel.org/r/20230102170759.29610-2-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Vivek has decided to transfer the maintainership of the VMware virtual
PTP clock driver (ptp_vmw) to Srivatsa and Deep. Update the
MAINTAINERS file to reflect this change, and also add Alexey as a
reviewer for the driver.
Signed-off-by: Srivatsa S. Bhat (VMware) <srivatsa@csail.mit.edu>
Acked-by: Vivek Thampi <vivek@vivekthampi.com>
Acked-by: Deep Shah <sdeep@vmware.com>
Acked-by: Alexey Makhalov <amakhalov@vmware.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Convert while loop into do-while with only a single call to
acpi_get_first_physical_node(). No functional change intended.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
The Qualcomm Soc cpufreq hardware engine has LMh/thermal throttling
interrupts (already present in SM8250 and SM8450 DTS) and Linux driver
uses them:
sm8250-hdk.dtb: cpufreq@18591000: 'interrupt-names', 'interrupts' do not match any of the regexes: 'pinctrl-[0-9]+'
sm8450-qrd.dtb: cpufreq@17d91000: 'interrupt-names', 'interrupts' do not match any of the regexes: 'pinctrl-[0-9]+'
Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
The Qualcomm SM6375 platform uses the qcom-cpufreq-hw driver, so add
it to the cpufreq-dt-platdev driver's blocklist.
Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Serialise access of TCP_Server_Info::hostname in
assemble_neg_contexts() by holding the server's mutex otherwise it
might end up accessing an already-freed hostname pointer from
cifs_reconnect() or cifs_resolve_server().
Signed-off-by: Paulo Alcantara (SUSE) <pc@cjr.nz>
Reviewed-by: Enzo Matsumiya <ematsumiya@suse.de>
Signed-off-by: Steve French <stfrench@microsoft.com>
If it failed to reconnect ipc used for getting referrals, we can just
ignore it as it is not required for reconnecting the share. The worst
case would be not being able to detect or chase nested links as long
as dfs root server is unreachable.
Before patch:
$ mount.cifs //root/dfs/link /mnt -o echo_interval=10,...
-> target share: /fs0/share
disconnect root & fs0
$ ls /mnt
ls: cannot access '/mnt': Host is down
connect fs0
$ ls /mnt
ls: cannot access '/mnt': Resource temporarily unavailable
After patch:
$ mount.cifs //root/dfs/link /mnt -o echo_interval=10,...
-> target share: /fs0/share
disconnect root & fs0
$ ls /mnt
ls: cannot access '/mnt': Host is down
connect fs0
$ ls /mnt
bar.rtf dir1 foo
Signed-off-by: Paulo Alcantara (SUSE) <pc@cjr.nz>
Reviewed-by: Enzo Matsumiya <ematsumiya@suse.de>
Signed-off-by: Steve French <stfrench@microsoft.com>
kmap_local_page() requires kunmap_local() to unmap the mapping. In
addition memcpy_page() is provided to perform this common memcpy
pattern.
Replace the kmap_local_page() and broken kunmap() with memcpy_page()
Fixes: d406d26745 ("cifs: skip alloc when request has no pages")
Reviewed-by: Paulo Alcantara <pc@cjr.nz>
Reviewed-by: "Fabio M. De Francesco" <fmdefrancesco@gmail.com>
Cc: linux-cifs@vger.kernel.org
Cc: samba-technical@lists.samba.org
Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
As reported by Viktor, plain accesses in LKMM are weaker than
accesses to registers: the latter carry dependencies but the former
do not. This is exemplified in the following snippet:
int r = READ_ONCE(*x);
WRITE_ONCE(*y, r);
Here a data dependency links the READ_ONCE() to the WRITE_ONCE(),
preserving their order, because the model treats r as a register.
If r is turned into a memory location accessed by plain accesses,
however, the link is broken and the order between READ_ONCE() and
WRITE_ONCE() is no longer preserved.
This is too conservative, since any optimizations on plain
accesses that might break dependencies are also possible on
registers; it also contradicts the intuitive notion of "dependency"
as the data stored by the WRITE_ONCE() does depend on the data read
by the READ_ONCE(), independently of whether r is a register or a
memory location.
This is resolved by redefining all dependencies to include
dependencies carried by memory accesses; a dependency is said to be
carried by memory accesses (in the model: carry-dep) from one load
to another load if the initial load is followed by an arbitrarily
long sequence alternating between stores and loads of the same
thread, where the data of each store depends on the previous load,
and is read by the next load.
Any dependency linking the final load in the sequence to another
access also links the initial load in the sequence to that access.
More deep details can be found in this LKML discussion:
https://lore.kernel.org/lkml/d86295788ad14a02874ab030ddb8a6f8@huawei.com/
Reported-by: Viktor Vafeiadis <viktor@mpi-sws.org>
Signed-off-by: Jonas Oberhauser <jonas.oberhauser@huawei.com>
Reviewed-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Viktor (as relayed by Jonas) has pointed out a weakness in the Linux
Kernel Memory Model. Namely, the memory ordering properties of atomic
operations are not monotonic: An atomic op with full-barrier semantics
does not always provide ordering as strong as one with release-barrier
semantics.
The following litmus test illustrates the problem:
--------------------------------------------------
C atomics-not-monotonic
{}
P0(int *x, atomic_t *y)
{
WRITE_ONCE(*x, 1);
smp_wmb();
atomic_set(y, 1);
}
P1(atomic_t *y)
{
int r1;
r1 = atomic_inc_return(y);
}
P2(int *x, atomic_t *y)
{
int r2;
int r3;
r2 = atomic_read(y);
smp_rmb();
r3 = READ_ONCE(*x);
}
exists (2:r2=2 /\ 2:r3=0)
--------------------------------------------------
The litmus test is allowed as shown with atomic_inc_return(), which
has full-barrier semantics. But if the operation is changed to
atomic_inc_return_release(), which only has release-barrier semantics,
the litmus test is forbidden. Clearly this violates monotonicity.
The reason is because the LKMM treats full-barrier atomic ops as if
they were written:
mb();
load();
store();
mb();
(where the load() and store() are the two parts of an atomic RMW op),
whereas it treats release-barrier atomic ops as if they were written:
load();
release_barrier();
store();
The difference is that here the release barrier orders the load part
of the atomic op before the store part with A-cumulativity, whereas
the mb()'s above do not. This means that release-barrier atomics can
effectively extend the cumul-fence relation but full-barrier atomics
cannot.
To resolve this problem we introduce the rmw-sequence relation,
representing an arbitrarily long sequence of atomic RMW operations in
which each operation reads from the previous one, and explicitly allow
it to extend cumul-fence. This modification of the memory model is
sound; it holds for PPC because of B-cumulativity, it holds for TSO
and ARM64 because of other-multicopy atomicity, and we can assume that
atomic ops on all other architectures will be implemented so as to
make it hold for them.
For similar reasons we also allow rmw-sequence to extend the
w-post-bounded relation, which is analogous to cumul-fence in some
ways.
Reported-by: Viktor Vafeiadis <viktor@mpi-sws.org>
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Reviewed-by: Jonas Oberhauser <jonas.oberhauser@huawei.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The cited commit describes that when using writel(), explicit wmb()
is not needed. wmb() is an expensive barrier. writel() uses the needed
platform specific barrier instead of wmb().
writeX() section of "KERNEL I/O BARRIER EFFECTS" already describes
ordering of I/O accessors with MMIO writes.
Hence add the comment for pseudo code of writel() and remove confusing
text around writel() and wmb().
commit 5846581e35 ("locking/memory-barriers.txt: Fix broken DMA vs. MMIO ordering example")
Co-developed-by: Will Deacon <will@kernel.org>
Signed-off-by: Will Deacon <will@kernel.org>
Signed-off-by: Parav Pandit <parav@nvidia.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
When cs_watchdog_read() is unable to get a qualifying clocksource read
within the limit set by max_cswd_read_retries, it prints a message
and marks the clocksource under test as unstable. But that message is
unclear to anyone unfamiliar with the code:
clocksource: timekeeping watchdog on CPU13: wd-tsc-wd read-back delay 1000614ns, attempt 3, marking unstable
Therefore, add some context so that the message appears as follows:
clocksource: timekeeping watchdog on CPU13: wd-tsc-wd excessive read-back delay of 1000614ns vs. limit of 125000ns, wd-wd read-back delay only 27ns, attempt 3, marking tsc unstable
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Cc: John Stultz <jstultz@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Stephen Boyd <sboyd@kernel.org>
Cc: Feng Tang <feng.tang@intel.com>
Currently, MAX_SKEW_USEC is set to 100 microseconds, which has worked
reasonably well. However, NTP is willing to tolerate 500 microseconds
of skew per second, and a clocksource that is good enough for NTP should
be good enough for the clocksource watchdog. The watchdog's skew is
controlled by MAX_SKEW_USEC and the CLOCKSOURCE_WATCHDOG_MAX_SKEW_US
Kconfig option. However, these values are doubled before being associated
with a clocksource's ->uncertainty_margin, and the ->uncertainty_margin
values of the pair of clocksource's being compared are summed before
checking against the skew.
Therefore, set both MAX_SKEW_USEC and the default for the
CLOCKSOURCE_WATCHDOG_MAX_SKEW_US Kconfig option to 125 microseconds of
skew per second, resulting in 500 microseconds of skew per second in
the clocksource watchdog's skew comparison.
Suggested-by Rik van Riel <riel@surriel.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Some "TSC fall back to HPET" messages appear on systems having more than
2 NUMA nodes:
clocksource: timekeeping watchdog on CPU168: hpet read-back delay of 4296200ns, attempt 4, marking unstable
The "hpet" here is misleading the clocksource watchdog is really
doing repeated reads of "hpet" in order to check for unrelated delays.
Therefore, print the name of the clocksource under test, prefixed by
"wd-" and suffixed by "-wd", for example, "wd-tsc-wd".
Signed-off-by: Yunying Sun <yunying.sun@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Allow translation of REPORT_SUPPORTED_OPERATION_CODES commands using
the command format 0x3, that is, checking support for commands that are
identified using an opcode and a service action.
Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
For SCSI ML byte:
In the case where a command is completed via libata EH:
irq -> ata_qc_complete() -> ata_qc_schedule_eh()
irq done
... -> ata_do_eh() -> ata_eh_link_autopsy() -> ata_eh_finish() ->
ata_eh_qc_complete() -> __ata_eh_qc_complete() -> __ata_qc_complete() ->
qc->complete_fn() (ata_scsi_qc_complete()) -> ata_qc_done() ->
qc->scsidone() (empty stub)
... -> scsi_eh_finish_cmd() -> scsi_eh_flush_done_q() ->
scsi_finish_command()
ata_eh_link_autopsy() will call ata_eh_analyze_tf(), which calls
scsi_check_sense(), which sets the SCSI ML byte.
Since ata_scsi_qc_complete() is called after scsi_check_sense() when
a command is completed via libata EH, we cannot simply overwrite the
SCSI ML byte that was set earlier in the call chain.
For SCSI status byte:
When a SCSI command is prepared using scsi_prepare_cmd(), it sets
cmd->result to 0. (SAM_STAT_GOOD is defined as 0x0).
Likewise, when a command is requeued from SCSI EH, scsi_queue_insert()
is called, which sets cmd->result to 0.
A SCSI command thus always has a GOOD status by default when being
sent to libata.
If libata fetches sense data from the device, it will call
ata_scsi_set_sense(), which will set the status byte to
SAM_STAT_CHECK_CONDITION, if the caller deems that the status should be
a check condition.
ata_scsi_qc_complete() should therefore never overwrite the existing
status byte, because if it is != GOOD, it was set by libata itself,
for a reason.
For the host byte:
When libata abort commands, because of a NCQ error, it will schedule
SCSI EH for all QCs using blk_abort_request(), which will all end up in
scsi_timeout(), which will call scsi_abort_command(). scsi_timeout()
sets DID_TIME_OUT regardless if a command was aborted or timed out.
If we don't clear the DID_TIME_OUT byte for the QC that caused the
NCQ error, that QC will be reported as a timed out command, instead
of being reported as a NCQ error.
For a command that actually timed out, DID_TIME_OUT would be fine to
keep, but libata has its own way of detecting that a command timed out
(see ata_scsi_cmd_error_handler()), and sets AC_ERR_TIMEOUT if that is
the case. libata will retry timed out commands.
We could clear DID_TIME_OUT only for the QC that caused the NCQ error,
but since libata has its own way of detecting timeouts, simply clear it
always.
Note that the existing ata_scsi_qc_complete() code does:
cmd->result = SAM_STAT_CHECK_CONDITION or cmd->result = SAM_STAT_GOOD.
This WILL clear the host byte. So us clearing the host byte
unconditionally is in line with the existing libata behavior.
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
ata_dev_configure() starts off by clearing all flags in ATA_DFLAG_CFG_MASK:
dev->flags &= ~ATA_DFLAG_CFG_MASK;
ata_dev_configure() then calls ata_dev_config_lba() which calls
ata_dev_config_ncq().
ata_dev_config_ncq() will set the correct ATA_DFLAGs depending on what is
actually supported.
Since these flags are set by ata_dev_configure(), they should be in
ATA_DFLAG_CFG_MASK and not in ATA_DFLAG_INIT_MASK.
ATA_DFLAG_NCQ_PRIO_ENABLED is set via sysfs, is should therefore not be in
ATA_DFLAG_CFG_MASK. It also cannot be in ATA_DFLAG_INIT_MASK, because
ata_eh_schedule_probe() calls ata_dev_init(), which will clear all flags in
ATA_DFLAG_INIT_MASK.
This means that ATA_DFLAG_NCQ_PRIO_ENABLED (the value the user sets via
sysfs) would get silently cleared if ata_eh_schedule_probe() is called.
While that should only happen in certain circumstances, it still doesn't
seem right that it can get silently cleared.
(ata_dev_config_ncq_prio() will still clear the ATA_DFLAG_NCQ_PRIO_ENABLED
flag if ATA_DFLAG_NCQ_PRIO is suddenly no longer supported after a
revalidation.)
Because of this, move ATA_DFLAG_NCQ_PRIO_ENABLED to be outside of both
ATA_DFLAG_CFG_MASK and ATA_DFLAG_INIT_MASK.
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
In AHCI specification 1.3.1:
"5.5.3 Processing Completed Commands"
"For each port that has an interrupt pending:
1. Software determines the cause of the interrupt by reading the PxIS
register. It is possible for multiple bits to be set.
2. Software clears appropriate bits in the PxIS register corresponding
to the cause of the interrupt.
3. Software clears the interrupt bit in IS.IPS corresponding to the port.
4. If executing non-queued commands, software reads the PxCI register,
and compares the current value to the list of commands previously
issued by software that are still outstanding. If executing native
queued commands, software reads the PxSACT register and compares the
current value to the list of commands previously issued by software.
Software completes with success any outstanding command whose
corresponding bit has been cleared in the respective register. PxCI
and PxSACT are volatile registers; software should only use their
values to determine commands that have completed, not to determine
which commands have previously been issued.
5. If there were errors, noted in the PxIS register, software performs
error recovery actions (see section 6.2.2)."
The documentation for the PxSACT shadow register in AHCI:
"The device clears bits in this field by sending a Set Device Bits FIS
to the host. The HBA clears bits in this field that are set to ‘1’ in
the SActive field of the Set Device Bits FIS. The HBA only clears bits
that correspond to native queued commands that have completed
successfully."
Additionally, in SATA specification 3.5a:
"11.15 FPDMA QUEUED command protocol"
"DFPDMAQ11: ERROR
Halt command processing and transmit Set Device Bits FIS to host
with the ERR bit in Status field set to one, Interrupt bit set to one,
ATA error code set to one in the ERROR field, bits in ACT field cleared
to zero for any outstanding queued commands, and bits set to one
for any successfully completed queued commands that completion
notification not yet delivered."
I.e. even when the HBA triggers an error interrupt, the HBA will still
clear successfully completed commands in PxSACT. Commands that did not
complete successfully will still have its bit set in PxSACT.
(Which means the command that caused the NCQ error and queued commands
that had not yet finished at the time when the NCQ error occurred.)
Additionally, for a HBA that does not have the libata flag
AHCI_HFLAG_MULTI_MSI set, all ap->locks will point to host->lock, which
means that IRQs will be disabled for one port while another port's IRQ
handler is running. The HBA will still receive FISes from the device,
even if IRQs on the HBA itself are disabled. What can thus e.g. receive
a FIS that completes several commands successfully, followed by a FIS
that does (or does not) complete additional commands with the error bit
set, to indicate that at least one command was aborted.
Therefore, modify ahci_handle_port_interrupt() using the new helper
ahci_qc_complete() to complete the commands that have already been
signaled as successfully through a regular completion SDB FIS, as not
doing so would simply cause successfully completed commands to be
retried for no good reason.
Co-developed-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Currently, the status is being read for each QC, inside
ata_qc_complete(), which means that QCs being completed by
ata_qc_complete_multiple() (i.e. multiple QCs completed during a single
interrupt), can have different status and error bits set. This is
because the FIS Receive Area will get updated as soon as the HBA
receives a new FIS from the device in the NCQ case.
Here is an example of the problem:
ata14.00: ata_qc_complete_multiple: done_mask: 0x180000
qc tag: 19 cmd: 0x61 flags: 0x11b err_mask: 0x0 tf->status: 0x40
qc tag: 20 cmd: 0x61 flags: 0x11b err_mask: 0x0 tf->status: 0x43
A print in ata_qc_complete_multiple(), shows that done_mask is: 0x180000
which means that tag 19 and 20 were completed. Another print in
ata_qc_complete(), after the call to fill_result_tf(), shows that tag 19
and 20 have different status values, even though they were completed in
the same ata_qc_complete_multiple() call.
If PMP is not enabled, simply read the status and error once, before
calling ata_qc_complete() for each QC. Without PMP, we know that all QCs
must share the same status and error values.
If PMP is enabled, we also read the status before calling
ata_qc_complete(), however, we still read the status for each QC, since
the QCs can belong to different PMP links (which means that the QCs
does not necessarily share the same status and error values).
Do all this by introducing the new port operation .qc_ncq_fill_rtf. If
set, this operation is called in ata_qc_complete_multiple() to set the
result tf for all completed QCs signaled by the last SDB FIS received.
QCs that have their result tf filled are marked with the new flag
ATA_QCFLAG_RTF_FILLED so that any later execution of the qc_fill_rtf
port operation does nothing (e.g. when called from ata_qc_complete()).
Co-developed-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
The boolean return value of the qc_fill_rtf operation is used nowhere.
Simplify this operation interface by making it a void function. All
drivers defining this operation are also updated.
Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Reviewed-by: John Garry <john.g.garry@oracle.com>
The name ATA_QCFLAG_FAILED is misleading since it does not mean that a
QC completed in error, or that it didn't complete at all. It means that
libata decided to schedule EH for the QC, so the QC is now owned by the
libata error handler (EH).
The normal execution path is responsible for not accessing a QC owned
by EH. libata core enforces the rule by returning NULL from
ata_qc_from_tag() for QCs owned by EH.
It is quite easy to mistake that a QC marked with ATA_QCFLAG_FAILED was
an error. However, a QC that was actually an error is instead indicated
by having qc->err_mask set. E.g. when we have a NCQ error, we abort all
QCs, which currently will mark all QCs as ATA_QCFLAG_FAILED. However, it
will only be a single QC that is an error (i.e. has qc->err_mask set).
Rename ATA_QCFLAG_FAILED to ATA_QCFLAG_EH to more clearly highlight that
this flag simply means that a QC is now owned by EH. This new name will
not mislead to think that the QC was an error (which is instead
indicated by having qc->err_mask set).
This also makes it more obvious that the EH code skips all QCs that do
not have ATA_QCFLAG_EH set (rather than ATA_QCFLAG_FAILED), since the EH
code should simply only care about QCs that are owned by EH itself.
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Reviewed-by: John Garry <john.g.garry@oracle.com>
Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
If ap->ops->error_handler is NULL just return. This patch also
fixes some comment style issue.
Signed-off-by: Wenchao Hao <haowenchao@huawei.com>
Reviewed-by: Niklas Cassel <niklas.cassel@wdc.com>
Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
There are 3 possible interrupt sources are handled by DP controller,
HPDstatus, Controller state changes and Aux read/write transaction.
At every irq, DP controller have to check isr status of every interrupt
sources and service the interrupt if its isr status bits shows interrupts
are pending. There is potential race condition may happen at current aux
isr handler implementation since it is always complete dp_aux_cmd_fifo_tx()
even irq is not for aux read or write transaction. This may cause aux read
transaction return premature if host aux data read is in the middle of
waiting for sink to complete transferring data to host while irq happen.
This will cause host's receiving buffer contains unexpected data. This
patch fixes this problem by checking aux isr and return immediately at
aux isr handler if there are no any isr status bits set.
Current there is a bug report regrading eDP edid corruption happen during
system booting up. After lengthy debugging to found that VIDEO_READY
interrupt was continuously firing during system booting up which cause
dp_aux_isr() to complete dp_aux_cmd_fifo_tx() prematurely to retrieve data
from aux hardware buffer which is not yet contains complete data transfer
from sink. This cause edid corruption.
Follows are the signature at kernel logs when problem happen,
EDID has corrupt header
panel-simple-dp-aux aux-aea0000.edp: Couldn't identify panel via EDID
Changes in v2:
-- do complete if (ret == IRQ_HANDLED) ay dp-aux_isr()
-- add more commit text
Changes in v3:
-- add Stephen suggested
-- dp_aux_isr() return IRQ_XXX back to caller
-- dp_ctrl_isr() return IRQ_XXX back to caller
Changes in v4:
-- split into two patches
Changes in v5:
-- delete empty line between tags
Changes in v6:
-- remove extra "that" and fixed line more than 75 char at commit text
Fixes: c943b4948b ("drm/msm/dp: add displayPort driver support")
Signed-off-by: Kuogee Hsieh <quic_khsieh@quicinc.com>
Tested-by: Douglas Anderson <dianders@chromium.org>
Reviewed-by: Abhinav Kumar <quic_abhinavk@quicinc.com>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
Patchwork: https://patchwork.freedesktop.org/patch/516121/
Link: https://lore.kernel.org/r/1672193785-11003-2-git-send-email-quic_khsieh@quicinc.com
Signed-off-by: Abhinav Kumar <quic_abhinavk@quicinc.com>
Unlike normal tw, nothing prevents deferred tw to be executed right
after an tw item added to ->work_llist in io_req_local_work_add(). For
instance, the waiting task may get waken up by CQ posting or a normal
tw. Thus we need to pin the ring for the rest of io_req_local_work_add()
Cc: stable@vger.kernel.org
Fixes: c0e0d6ba25 ("io_uring: add IORING_SETUP_DEFER_TASKRUN")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/1a79362b9c10b8523ef70b061d96523650a23344.1672795998.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This commit upgrades the kvm.sh script's --kconfig parameter to accept
string-valued Kconfig options with double-quoted string values.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently, the presence of any quoted-string Kconfig option in the
scenario files or the CFcommon file (aside from the special-cased
CONFIG_INITRAMFS_SOURCE option) will result in an "improperly set"
diagnostic. This commit updates configcheck.sh to strip double quotes
in order to permit string-valued Kconfig options to be handled correctly.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The latest version of grep is deprecating the egrep command, so that
its output contains warnings as follows:
egrep: warning: egrep is obsolescent; using grep -E
Fix this using "grep -E" instead.
sed -i "s/egrep/grep -E/g" `grep egrep -rwl tools/testing/selftests/rcutorture`
Here are the steps to install the latest grep:
wget http://ftp.gnu.org/gnu/grep/grep-3.8.tar.gz
tar xf grep-3.8.tar.gz
cd grep-3.8 && ./configure && make
sudo make install
export PATH=/usr/local/bin:$PATH
Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Under some conditions, a given run's vmlinux file will be compressed,
so that it is named vmlinux.xz rather than vmlinux. in such cases,
kvm-find-errors.sh will complain about the nonexistence of vmlinux.
This commit therefore causes kvm-find-errors.sh to check for vmlinux.xz
as well as for vmlinux.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Current tests all have init() functions that are guaranteed to succeed.
But upcoming tests will need to allocate memory, thus possibly failing.
This commit therefore handles init() function failure.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The DEFINE_TORTURE_RANDOM_PERCPU() macro defines per-CPU random-number
generators for torture testing, but the seeds for each CPU's instance
will be identical if they are first used at the same time. This commit
therefore adds the CPU number to the mix when reseeding.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The rcu_tasks_need_gpcb() determines whether or not: (1) There are
callbacks needing another grace period, (2) There are callbacks ready
to be invoked, and (3) It would be a good time to shrink back down to a
single-CPU callback list. This third case is interesting because some
other CPU might be adding new callbacks, which might suddenly make this
a very bad time to be shrinking.
This is currently handled by requiring call_rcu_tasks_generic() to
enqueue callbacks under the protection of rcu_read_lock() and requiring
rcu_tasks_need_gpcb() to wait for an RCU grace period to elapse before
finalizing the transition. This works well in practice.
Unfortunately, the current code assumes that a grace period whose end is
detected by the poll_state_synchronize_rcu() in the second "if" condition
actually ended before the earlier code counted the callbacks queued on
CPUs other than CPU 0 (local variable "ncbsnz"). Given the current code,
it is possible that a long-delayed call_rcu_tasks_generic() invocation
will queue a callback on a non-zero CPU after these CPUs have had their
callbacks counted and zero has been stored to ncbsnz. Such a callback
would trigger the WARN_ON_ONCE() in the second "if" statement.
To see this, consider the following sequence of events:
o CPU 0 invokes rcu_tasks_one_gp(), and counts fewer than
rcu_task_collapse_lim callbacks. It sees at least one
callback queued on some other CPU, thus setting ncbsnz
to a non-zero value.
o CPU 1 invokes call_rcu_tasks_generic() and loads 42 from
->percpu_enqueue_lim. It therefore decides to enqueue its
callback onto CPU 1's callback list, but is delayed.
o CPU 0 sees the rcu_task_cb_adjust is non-zero and that the number
of callbacks does not exceed rcu_task_collapse_lim. It therefore
checks percpu_enqueue_lim, and sees that its value is greater
than the value one. CPU 0 therefore starts the shift back
to a single callback list. It sets ->percpu_enqueue_lim to 1,
but CPU 1 has already read the old value of 42. It also gets
a grace-period state value from get_state_synchronize_rcu().
o CPU 0 sees that ncbsnz is non-zero in its second "if" statement,
so it declines to finalize the shrink operation.
o CPU 0 again invokes rcu_tasks_one_gp(), and counts fewer than
rcu_task_collapse_lim callbacks. It also sees that there are
no callback queued on any other CPU, and thus sets ncbsnz to zero.
o CPU 1 resumes execution and enqueues its callback onto its own
list. This invalidates the value of ncbsnz.
o CPU 0 sees the rcu_task_cb_adjust is non-zero and that the number
of callbacks does not exceed rcu_task_collapse_lim. It therefore
checks percpu_enqueue_lim, but sees that its value is already
unity. It therefore does not get a new grace-period state value.
o CPU 0 sees that rcu_task_cb_adjust is non-zero, ncbsnz is zero,
and that poll_state_synchronize_rcu() says that the grace period
has completed. it therefore finalizes the shrink operation,
setting ->percpu_dequeue_lim to the value one.
o CPU 0 does a debug check, scanning the other CPUs' callback lists.
It sees that CPU 1's list has a callback, so it (rightly)
triggers the WARN_ON_ONCE(). After all, the new value of
->percpu_dequeue_lim says to not bother looking at CPU 1's
callback list, which means that this callback will never be
invoked. This can result in hangs and maybe even OOMs.
Based on long experience with rcutorture, this is an extremely
low-probability race condition, but it really can happen, especially in
preemptible kernels or within guest OSes.
This commit therefore checks for completion of the grace period
before counting callbacks. With this change, in the above failure
scenario CPU 0 would know not to prematurely end the shrink operation
because the grace period would not have completed before the count
operation started.
[ paulmck: Adjust grace-period end rather than adding RCU reader. ]
[ paulmck: Avoid spurious WARN_ON_ONCE() with ->percpu_dequeue_lim check. ]
Signed-off-by: Zqiang <qiang1.zhang@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The synchronize_rcu_tasks_rude() function invokes rcu_tasks_rude_wait_gp()
to wait one rude RCU-tasks grace period. The rcu_tasks_rude_wait_gp()
function in turn checks if there is only a single online CPU. If so, it
will immediately return, because a call to synchronize_rcu_tasks_rude()
is by definition a grace period on a single-CPU system. (We could
have blocked!)
Unfortunately, this check uses num_online_cpus() without synchronization,
which can result in too-short grace periods. To see this, consider the
following scenario:
CPU0 CPU1 (going offline)
migration/1 task:
cpu_stopper_thread
-> take_cpu_down
-> _cpu_disable
(dec __num_online_cpus)
->cpuhp_invoke_callback
preempt_disable
access old_data0
task1
del old_data0 .....
synchronize_rcu_tasks_rude()
task1 schedule out
....
task2 schedule in
rcu_tasks_rude_wait_gp()
->__num_online_cpus == 1
->return
....
task1 schedule in
->free old_data0
preempt_enable
When CPU1 decrements __num_online_cpus, its value becomes 1. However,
CPU1 has not finished going offline, and will take one last trip through
the scheduler and the idle loop before it actually stops executing
instructions. Because synchronize_rcu_tasks_rude() is mostly used for
tracing, and because both the scheduler and the idle loop can be traced,
this means that CPU0's prematurely ended grace period might disrupt the
tracing on CPU1. Given that this disruption might include CPU1 executing
instructions in memory that was just now freed (and maybe reallocated),
this is a matter of some concern.
This commit therefore removes that problematic single-CPU check from the
rcu_tasks_rude_wait_gp() function. This dispenses with the single-CPU
optimization, but there is no evidence indicating that this optimization
is important. In addition, synchronize_rcu_tasks_generic() contains a
similar optimization (albeit only for early boot), which also splats.
(As in exactly why are you invoking synchronize_rcu_tasks_rude() so
early in boot, anyway???)
It is OK for the synchronize_rcu_tasks_rude() function's check to be
unsynchronized because the only times that this check can evaluate to
true is when there is only a single CPU running with preemption
disabled.
While in the area, this commit also fixes a minor bug in which a
call to synchronize_rcu_tasks_rude() would instead be attributed to
synchronize_rcu_tasks().
[ paulmck: Add "synchronize_" prefix and "()" suffix. ]
Signed-off-by: Zqiang <qiang1.zhang@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
RCU Tasks and PID-namespace unshare can interact in do_exit() in a
complicated circular dependency:
1) TASK A calls unshare(CLONE_NEWPID), this creates a new PID namespace
that every subsequent child of TASK A will belong to. But TASK A
doesn't itself belong to that new PID namespace.
2) TASK A forks() and creates TASK B. TASK A stays attached to its PID
namespace (let's say PID_NS1) and TASK B is the first task belonging
to the new PID namespace created by unshare() (let's call it PID_NS2).
3) Since TASK B is the first task attached to PID_NS2, it becomes the
PID_NS2 child reaper.
4) TASK A forks() again and creates TASK C which get attached to PID_NS2.
Note how TASK C has TASK A as a parent (belonging to PID_NS1) but has
TASK B (belonging to PID_NS2) as a pid_namespace child_reaper.
5) TASK B exits and since it is the child reaper for PID_NS2, it has to
kill all other tasks attached to PID_NS2, and wait for all of them to
die before getting reaped itself (zap_pid_ns_process()).
6) TASK A calls synchronize_rcu_tasks() which leads to
synchronize_srcu(&tasks_rcu_exit_srcu).
7) TASK B is waiting for TASK C to get reaped. But TASK B is under a
tasks_rcu_exit_srcu SRCU critical section (exit_notify() is between
exit_tasks_rcu_start() and exit_tasks_rcu_finish()), blocking TASK A.
8) TASK C exits and since TASK A is its parent, it waits for it to reap
TASK C, but it can't because TASK A waits for TASK B that waits for
TASK C.
Pid_namespace semantics can hardly be changed at this point. But the
coverage of tasks_rcu_exit_srcu can be reduced instead.
The current task is assumed not to be concurrently reapable at this
stage of exit_notify() and therefore tasks_rcu_exit_srcu can be
temporarily relaxed without breaking its constraints, providing a way
out of the deadlock scenario.
[ paulmck: Fix build failure by adding additional declaration. ]
Fixes: 3f95aa81d2 ("rcu: Make TASKS_RCU handle tasks that are almost done exiting")
Reported-by: Pengfei Xu <pengfei.xu@intel.com>
Suggested-by: Boqun Feng <boqun.feng@gmail.com>
Suggested-by: Neeraj Upadhyay <quic_neeraju@quicinc.com>
Suggested-by: Paul E. McKenney <paulmck@kernel.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Cc: Eric W . Biederman <ebiederm@xmission.com>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Ever since the following commit:
5a41344a3d ("srcu: Simplify __srcu_read_unlock() via this_cpu_dec()")
SRCU doesn't rely anymore on preemption to be disabled in order to
modify the per-CPU counter. And even then it used to be done from the API
itself.
Therefore and after checking further, it appears to be safe to remove
the preemption disablement around __srcu_read_[un]lock() in
exit_tasks_rcu_start() and exit_tasks_rcu_finish()
Suggested-by: Boqun Feng <boqun.feng@gmail.com>
Suggested-by: Paul E. McKenney <paulmck@kernel.org>
Suggested-by: Neeraj Upadhyay <quic_neeraju@quicinc.com>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
