Commit 6879c042e1 ("tools/bpf: selftests: Add bpf_iter selftests")
added self tests for bpf_iter feature. But two subtests
ipv6_route and netlink needs llvm latest 10.x release branch
or trunk due to a bug in llvm BPF backend. This patch added
the file README.rst to document these two failures
so people using llvm 10.0.0 can be aware of them.
Suggested-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200513180215.2949237-1-yhs@fb.com
Martin Blumenstingl says:
====================
dwmac-meson8b Ethernet RX delay configuration
The Ethernet TX performance has been historically bad on Meson8b and
Meson8m2 SoCs because high packet loss was seen. I found out that this
was related (yet again) to the RGMII TX delay configuration.
In the process of discussing the big picture (and not just a single
patch) [0] with Andrew I discovered that the IP block behind the
dwmac-meson8b driver actually seems to support the configuration of the
RGMII RX delay (at least on the Meson8b SoC generation).
Since I sent the first RFC I got additional documentation from Jianxin
(many thanks!). Also I have discovered some more interesting details:
- Meson8b Odroid-C1 requires an RX delay (by either the PHY or the MAC)
Based on the vendor u-boot code (not upstream) I assume that it will
be the same for all Meson8b and Meson8m2 boards
- Khadas VIM2 seems to have the RX delay built into the PCB trace
length. When I enable the RX delay on the PHY or MAC I can't get any
data through. I expect that we will have the same situation on all
GXBB, GXM, AXG, G12A, G12B and SM1 boards. Further clarification is
needed here though (since I can't visually see these lengthened
traces on the PCB). This will be done before sending patches for
these boards.
Dependencies for this series:
There is a soft dependency for patch #2 on commit f22531438f
"dt-bindings: net: dwmac: increase 'maxItems' for 'clocks',
'clock-names' properties" which is currently in Rob's -next tree.
That commit is needed to make the dt-bindings schema validation
pass for patch #2. That patch has been for ~4 weeks in Robs tree,
so I assume that is not going to be dropped.
Changes since RFC v2 at [2]:
- dropped $ref: /schemas/types.yaml#definitions/uint32 from the
"amlogic,rx-delay-ns" in patch #1 ("Don't need to define the
type when in standard units." says Rob - thanks, I learned
something new). Also use "default: 0" for for this property
instead of explaining it in the description text.
- added a note to the cover-letter about a hidden dependency for
dt-binding schema validation in patch #2
- Added Andrew's Reviewed-by to patches 1-7. Thank you again for
the quick and detailed reviews, I appreciate this!
- error out if the (optional) timing-adjustment clock is missing
but we're asked to enable the RGMII RX delay. The MAC won't
work in this specific case and either the RX delay has to be
provided by the PHY or the timing-adjustment clock has to be
added.
- dropped the dts patches (#9-11) which were only added to give
an overview how this is going to be used. those will be sent
separately
- dropped the RFC prefix
Changes since RFC v1 at [1]:
- add support for the timing adjustment clock input (dt-bindings and
in the driver) thanks to the input from the unnamed Ethernet engineer
at Amlogic. This is the missing link between the fclk_div2 clock and
the Ethernet controller on Meson8b (no traffic would flow if that
clock was disabled)
- add support fot the amlogic,rx-delay-ns property. The only supported
values so far are 0ns and 2ns. The registers seem to allow more
precise timing adjustments, but I could not make that work so far.
- add more register documentation (for the new RX delay bits) and
unified the placement of existing register documentation. Again,
thanks to Jianxin and the unnamed Ethernet engineer at Amlogic
- DO NOT MERGE: .dts patches to show the conversion of the Meson8b
and Meson8m2 boards to "rgmii-id". I didn't have time for all arm64
patches yet, but these will switch to phy-mode = "rgmii-txid" with
amlogic,rx-delay-ns = <0> (because the delay seems to be provided by
the PCB trace length).
[0] https://patchwork.kernel.org/patch/11309891/
[1] https://patchwork.kernel.org/cover/11310719/
[2] https://patchwork.kernel.org/cover/11518257/
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Configure the PRG_ETH0_ADJ_* bits to enable or disable the RX delay
based on the various RGMII PHY modes. For now the only supported RX
delay settings are:
- disabled, use for example for phy-mode "rgmii-id"
- 0ns - this is treated identical to "disabled", used for example on
boards where the PHY provides 2ns TX delay and the PCB trace length
already adds 2ns RX delay
- 2ns - for whenever the PHY cannot add the RX delay and the traces on
the PCB don't add any RX delay
Disabling the RX delay (in case u-boot enables it, which is the case
for example on Meson8b Odroid-C1) simply means that PRG_ETH0_ADJ_ENABLE,
PRG_ETH0_ADJ_SETUP, PRG_ETH0_ADJ_DELAY and PRG_ETH0_ADJ_SKEW should be
disabled (just disabling PRG_ETH0_ADJ_ENABLE may be enough, since that
disables the whole re-timing logic - but I find it makes more sense to
clear the other bits as well since they depend on that setting).
u-boot on Odroid-C1 uses the following steps to enable a 2ns RX delay:
- enabling enabling the timing adjustment clock
- enabling the timing adjustment logic by setting PRG_ETH0_ADJ_ENABLE
- setting the PRG_ETH0_ADJ_SETUP bit
The documentation for the PRG_ETH0_ADJ_DELAY and PRG_ETH0_ADJ_SKEW
registers indicates that we can even set different RX delays. However,
I could not find out how this works exactly, so for now we only support
a 2ns RX delay using the exact same way that Odroid-C1's u-boot does.
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The timing adjustment clock will need similar logic as the RGMII clock:
It has to be enabled in the driver conditionally and when the driver is
unloaded it should be disabled again. Extract the existing code for the
RGMII clock into a new function so it can be re-used.
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The PRG_ETHERNET registers have a built-in timing adjustment circuit
which can provide the RX delay in RGMII mode. This is driven by an
external (to this IP, but internal to the SoC) clock input. Fetch this
clock as optional (even though it's there on all supported SoCs) since
we just learned about it and existing .dtbs don't specify it.
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The PRG_ETH0_ADJ_* are used for applying the RGMII RX delay. The public
datasheets only have very limited description for these registers, but
Jianxin Pan provided more detailed documentation from an (unnamed)
Amlogic engineer. Add the PRG_ETH0_ADJ_* bits along with the improved
description.
Suggested-by: Jianxin Pan <jianxin.pan@amlogic.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move the documentation for the TX delay above the PRG_ETH0_TXDLY_MASK
definition. Future commits will add more registers also with
documentation above their register bit definitions. Move the existing
comment so it will be consistent with the upcoming changes.
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use FIELD_PREP() to shift a value to the correct offset based on a
bitmask instead of open-coding the logic.
No functional changes.
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The PRG_ETHERNET registers can add an RX delay in RGMII mode. This
requires an internal re-timing circuit whose input clock is called
"timing adjustment clock". Document this clock input so the clock can be
enabled as needed.
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The PRG_ETHERNET registers on Meson8b and newer SoCs can add an RX
delay. Add a property with the known supported values so it can be
configured according to the board layout.
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Johan Hedberg says:
====================
pull request: bluetooth-next 2020-05-13
Here's a second attempt at a bluetooth-next pull request which
supercedes the one dated 2020-05-09. This should have the issues
discovered by Jakub fixed.
- Add support for Intel Typhoon Peak device (8087:0032)
- Add device tree bindings for Realtek RTL8723BS device
- Add device tree bindings for Qualcomm QCA9377 device
- Add support for experimental features configuration through mgmt
- Add driver hook to prevent wake from suspend
- Add support for waiting for L2CAP disconnection response
- Multiple fixes & cleanups to the btbcm driver
- Add support for LE scatternet topology for selected devices
- A few other smaller fixes & cleanups
Please let me know if there are any issues pulling. Thanks.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Andrii Nakryiko says:
====================
Add generic benchmark runner framework which simplifies writing various
performance benchmarks in a consistent fashion. This framework will be used
in follow up patches to test performance of perf buffer and ring buffer as
well.
Patch #1 extracts parse_num_list to be re-used between test_progs and bench.
Patch #2 adds generic runner implementation and atomic counter benchmarks to
validate benchmark runner's behavior.
Patch #3 implements test_overhead benchmark as part of bench runner. It also
add fmod_ret BPF program type to a set of benchmarks.
Patch #4 tests faster alternatives to set_task_comm() approach, tested in
test_overhead, in search for minimal-overhead way to trigger BPF program
execution from user-space on demand.
v2->v3:
- added --prod-affinity and --cons-affinity (Yonghong);
- removed ringbuf-related options leftovers (Yonghong);
- added more benchmarking results for test_overhead performance discrepancies;
v1->v2:
- moved benchmarks into benchs/ subdir (John);
- added benchmark "suite" scripts (John);
- few small clean ups, change defaults, etc.
====================
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
It is sometimes desirable to be able to trigger BPF program from user-space
with minimal overhead. sys_enter would seem to be a good candidate, yet in
a lot of cases there will be a lot of noise from syscalls triggered by other
processes on the system. So while searching for low-overhead alternative, I've
stumbled upon getpgid() syscall, which seems to be specific enough to not
suffer from accidental syscall by other apps.
This set of benchmarks compares tp, raw_tp w/ filtering by syscall ID, kprobe,
fentry and fmod_ret with returning error (so that syscall would not be
executed), to determine the lowest-overhead way. Here are results on my
machine (using benchs/run_bench_trigger.sh script):
base : 9.200 ± 0.319M/s
tp : 6.690 ± 0.125M/s
rawtp : 8.571 ± 0.214M/s
kprobe : 6.431 ± 0.048M/s
fentry : 8.955 ± 0.241M/s
fmodret : 8.903 ± 0.135M/s
So it seems like fmodret doesn't give much benefit for such lightweight
syscall. Raw tracepoint is pretty decent despite additional filtering logic,
but it will be called for any other syscall in the system, which rules it out.
Fentry, though, seems to be adding the least amoung of overhead and achieves
97.3% of performance of baseline no-BPF-attached syscall.
Using getpgid() seems to be preferable to set_task_comm() approach from
test_overhead, as it's about 2.35x faster in a baseline performance.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20200512192445.2351848-5-andriin@fb.com
Add fmod_ret BPF program to existing test_overhead selftest. Also re-implement
user-space benchmarking part into benchmark runner to compare results. Results
with ./bench are consistently somewhat lower than test_overhead's, but relative
performance of various types of BPF programs stay consisten (e.g., kretprobe is
noticeably slower). This slowdown seems to be coming from the fact that
test_overhead is single-threaded, while benchmark always spins off at least
one thread for producer. This has been confirmed by hacking multi-threaded
test_overhead variant and also single-threaded bench variant. Resutls are
below. run_bench_rename.sh script from benchs/ subdirectory was used to
produce results for ./bench.
Single-threaded implementations
===============================
/* bench: single-threaded, atomics */
base : 4.622 ± 0.049M/s
kprobe : 3.673 ± 0.052M/s
kretprobe : 2.625 ± 0.052M/s
rawtp : 4.369 ± 0.089M/s
fentry : 4.201 ± 0.558M/s
fexit : 4.309 ± 0.148M/s
fmodret : 4.314 ± 0.203M/s
/* selftest: single-threaded, no atomics */
task_rename base 4555K events per sec
task_rename kprobe 3643K events per sec
task_rename kretprobe 2506K events per sec
task_rename raw_tp 4303K events per sec
task_rename fentry 4307K events per sec
task_rename fexit 4010K events per sec
task_rename fmod_ret 3984K events per sec
Multi-threaded implementations
==============================
/* bench: multi-threaded w/ atomics */
base : 3.910 ± 0.023M/s
kprobe : 3.048 ± 0.037M/s
kretprobe : 2.300 ± 0.015M/s
rawtp : 3.687 ± 0.034M/s
fentry : 3.740 ± 0.087M/s
fexit : 3.510 ± 0.009M/s
fmodret : 3.485 ± 0.050M/s
/* selftest: multi-threaded w/ atomics */
task_rename base 3872K events per sec
task_rename kprobe 3068K events per sec
task_rename kretprobe 2350K events per sec
task_rename raw_tp 3731K events per sec
task_rename fentry 3639K events per sec
task_rename fexit 3558K events per sec
task_rename fmod_ret 3511K events per sec
/* selftest: multi-threaded, no atomics */
task_rename base 3945K events per sec
task_rename kprobe 3298K events per sec
task_rename kretprobe 2451K events per sec
task_rename raw_tp 3718K events per sec
task_rename fentry 3782K events per sec
task_rename fexit 3543K events per sec
task_rename fmod_ret 3526K events per sec
Note that the fact that ./bench benchmark always uses atomic increments for
counting, while test_overhead doesn't, doesn't influence test results all that
much.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20200512192445.2351848-4-andriin@fb.com
While working on BPF ringbuf implementation, testing, and benchmarking, I've
developed a pretty generic and modular benchmark runner, which seems to be
generically useful, as I've already used it for one more purpose (testing
fastest way to trigger BPF program, to minimize overhead of in-kernel code).
This patch adds generic part of benchmark runner and sets up Makefile for
extending it with more sets of benchmarks.
Benchmarker itself operates by spinning up specified number of producer and
consumer threads, setting up interval timer sending SIGALARM signal to
application once a second. Every second, current snapshot with hits/drops
counters are collected and stored in an array. Drops are useful for
producer/consumer benchmarks in which producer might overwhelm consumers.
Once test finishes after given amount of warm-up and testing seconds, mean and
stddev are calculated (ignoring warm-up results) and is printed out to stdout.
This setup seems to give consistent and accurate results.
To validate behavior, I added two atomic counting tests: global and local.
For global one, all the producer threads are atomically incrementing same
counter as fast as possible. This, of course, leads to huge drop of
performance once there is more than one producer thread due to CPUs fighting
for the same memory location.
Local counting, on the other hand, maintains one counter per each producer
thread, incremented independently. Once per second, all counters are read and
added together to form final "counting throughput" measurement. As expected,
such setup demonstrates linear scalability with number of producers (as long
as there are enough physical CPU cores, of course). See example output below.
Also, this setup can nicely demonstrate disastrous effects of false sharing,
if care is not taken to take those per-producer counters apart into
independent cache lines.
Demo output shows global counter first with 1 producer, then with 4. Both
total and per-producer performance significantly drop. The last run is local
counter with 4 producers, demonstrating near-perfect scalability.
$ ./bench -a -w1 -d2 -p1 count-global
Setting up benchmark 'count-global'...
Benchmark 'count-global' started.
Iter 0 ( 24.822us): hits 148.179M/s (148.179M/prod), drops 0.000M/s
Iter 1 ( 37.939us): hits 149.308M/s (149.308M/prod), drops 0.000M/s
Iter 2 (-10.774us): hits 150.717M/s (150.717M/prod), drops 0.000M/s
Iter 3 ( 3.807us): hits 151.435M/s (151.435M/prod), drops 0.000M/s
Summary: hits 150.488 ± 1.079M/s (150.488M/prod), drops 0.000 ± 0.000M/s
$ ./bench -a -w1 -d2 -p4 count-global
Setting up benchmark 'count-global'...
Benchmark 'count-global' started.
Iter 0 ( 60.659us): hits 53.910M/s ( 13.477M/prod), drops 0.000M/s
Iter 1 (-17.658us): hits 53.722M/s ( 13.431M/prod), drops 0.000M/s
Iter 2 ( 5.865us): hits 53.495M/s ( 13.374M/prod), drops 0.000M/s
Iter 3 ( 0.104us): hits 53.606M/s ( 13.402M/prod), drops 0.000M/s
Summary: hits 53.608 ± 0.113M/s ( 13.402M/prod), drops 0.000 ± 0.000M/s
$ ./bench -a -w1 -d2 -p4 count-local
Setting up benchmark 'count-local'...
Benchmark 'count-local' started.
Iter 0 ( 23.388us): hits 640.450M/s (160.113M/prod), drops 0.000M/s
Iter 1 ( 2.291us): hits 605.661M/s (151.415M/prod), drops 0.000M/s
Iter 2 ( -6.415us): hits 607.092M/s (151.773M/prod), drops 0.000M/s
Iter 3 ( -1.361us): hits 601.796M/s (150.449M/prod), drops 0.000M/s
Summary: hits 604.849 ± 2.739M/s (151.212M/prod), drops 0.000 ± 0.000M/s
Benchmark runner supports setting thread affinity for producer and consumer
threads. You can use -a flag for default CPU selection scheme, where first
consumer gets CPU #0, next one gets CPU #1, and so on. Then producer threads
pick up next CPU and increment one-by-one as well. But user can also specify
a set of CPUs independently for producers and consumers with --prod-affinity
1,2-10,15 and --cons-affinity <set-of-cpus>. The latter allows to force
producers and consumers to share same set of CPUs, if necessary.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/bpf/20200512192445.2351848-3-andriin@fb.com
Ursula Braun says:
====================
s390/net: updates 2020-05-13
please apply the fix from Wei Yongjun to netdev's net tree and
add Karsten Graul as co-maintainer for drivers/s390/net.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Add Karsten as additional maintainer for drivers/s390/net .
One of his focal points is the ism driver.
Cc: Julian Wiedmann <jwi@linux.ibm.com>
Acked-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix to return negative error code -ENOMEM from the smcd_alloc_dev()
error handling case instead of 0, as done elsewhere in this function.
Fixes: 684b89bc39 ("s390/ism: add device driver for internal shared memory")
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The current codebase makes use of the zero-length array language
extension to the C90 standard, but the preferred mechanism to declare
variable-length types such as these ones is a flexible array member[1][2],
introduced in C99:
struct foo {
int stuff;
struct boo array[];
};
By making use of the mechanism above, we will get a compiler warning
in case the flexible array does not occur last in the structure, which
will help us prevent some kind of undefined behavior bugs from being
inadvertently introduced[3] to the codebase from now on.
Also, notice that, dynamic memory allocations won't be affected by
this change:
"Flexible array members have incomplete type, and so the sizeof operator
may not be applied. As a quirk of the original implementation of
zero-length arrays, sizeof evaluates to zero."[1]
sizeof(flexible-array-member) triggers a warning because flexible array
members have incomplete type[1]. There are some instances of code in
which the sizeof operator is being incorrectly/erroneously applied to
zero-length arrays and the result is zero. Such instances may be hiding
some bugs. So, this work (flexible-array member conversions) will also
help to get completely rid of those sorts of issues.
This issue was found with the help of Coccinelle.
[1] https://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html
[2] https://github.com/KSPP/linux/issues/21
[3] commit 7649773293 ("cxgb3/l2t: Fix undefined behaviour")
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: Song Liu <songliubraving@fb.com>
When we changed the file registration handling, it became important to
iterate the bulk request freeing list for fixed files as well, or we
miss dropping the fixed file reference. If not, we're leaking references,
and we'll get a kworker stuck waiting for file references to disappear.
This also means we can remove the special casing of fixed vs non-fixed
files, we need to iterate for both and we can just rely on
__io_req_aux_free() doing io_put_file() instead of doing it manually.
Fixes: 0558955373 ("io_uring: refactor file register/unregister/update handling")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Xiaoliang Yang says:
====================
net: dsa: felix: tc taprio and CBS offload support
This patch series support tc taprio and CBS hardware offload according
to IEEE 802.1Qbv and IEEE-802.1Qav on VSC9959.
v1->v2 changes:
- Move port_qos_map_init() function to be common felix codes.
- Keep const for dsa_switch_ops structs, add felix_port_setup_tc
function to call port_setup_tc of felix.info.
- fix code style for cbs_set, rename variables.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
VSC9959 hardware support the Credit Based Shaper(CBS) which part
of the IEEE-802.1Qav. This patch support sch_cbs set for VSC9959.
Signed-off-by: Xiaoliang Yang <xiaoliang.yang_1@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Ocelot VSC9959 switch supports time-based egress shaping in hardware
according to IEEE 802.1Qbv. This patch add support for TAS configuration
on egress port of VSC9959 switch.
Felix driver is an instance of Ocelot family, with a DSA front-end. The
patch uses tc taprio hardware offload to setup TAS set function on felix
driver.
Signed-off-by: Xiaoliang Yang <xiaoliang.yang_1@nxp.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Set the default QoS Classification based on PCP and DEI of vlan tag,
after that, frames can be Classified to different Qos based on PCP tag.
If there is no vlan tag or vlan ignored, use port default Qos.
Signed-off-by: Xiaoliang Yang <xiaoliang.yang_1@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
From the mlx5-next branch at
git://git.kernel.org/pub/scm/linux/kernel/git/mellanox/linux
Required for dependencies in following patches
* branch 'mellanox/mlx5-next':
net/mlx5: Add support in forward to namespace
{IB/net}/mlx5: Simplify don't trap code
net/mlx5: Replace zero-length array with flexible-array
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
When using RAID1 and write-behind, md can deadlock when errors occur. With
write-behind, r1bio structs can be accounted by raid1 as queued but not
counted as pending. The pending count is dropped when the original bio is
returned complete but write-behind for the r1bio may still be active.
This breaks the accounting used in some conditions to know when the raid1
md device has reached an idle state. It can result in calls to
freeze_array deadlocking. freeze_array will never complete from a negative
"unqueued" value being calculated due to a queued count larger than the
pending count.
To properly account for write-behind, move the call to allow_barrier from
call_bio_endio to raid_end_bio_io. When using write-behind, md can call
call_bio_endio before all write-behind I/O is complete. Using
raid_end_bio_io for the point to call allow_barrier will release the
pending count at a point where all I/O for an r1bio, even write-behind, is
done.
Signed-off-by: David Jeffery <djeffery@redhat.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
In mddev_create_serial_pool(), memalloc scope APIs memalloc_noio_save()
and memalloc_noio_restore() are used when allocating memory by calling
mempool_create_kmalloc_pool(). After adding the memalloc scope APIs in
raid array suspend context, it is unncessary to explicitly call them
around mempool_create_kmalloc_pool() any longer.
This patch removes the redundant memalloc scope APIs in
mddev_create_serial_pool().
Signed-off-by: Coly Li <colyli@suse.de>
Cc: Guoqing Jiang <guoqing.jiang@cloud.ionos.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
Code comments of scribble_alloc() is outdated for a while. This patch
update the comments in function header for the new parameter list.
Suggested-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Song Liu <songliubraving@fb.com>
Using GFP_NOIO flag to call scribble_alloc() from resize_chunk() does
not have the expected behavior. kvmalloc_array() inside scribble_alloc()
which receives the GFP_NOIO flag will eventually call kmalloc_node() to
allocate physically continuous pages.
Now we have memalloc scope APIs in mddev_suspend()/mddev_resume() to
prevent memory reclaim I/Os during raid array suspend context, calling
to kvmalloc_array() with GFP_KERNEL flag may avoid deadlock of recursive
I/O as expected.
This patch removes the useless gfp flags from parameters list of
scribble_alloc(), and call kvmalloc_array() with GFP_KERNEL flag. The
incorrect GFP_NOIO flag does not exist anymore.
Fixes: b330e6a49d ("md: convert to kvmalloc")
Suggested-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Song Liu <songliubraving@fb.com>
In raid5.c:resize_chunk(), scribble_alloc() is called with GFP_NOIO
flag, then it is sent into kvmalloc_array() inside scribble_alloc().
The problem is kvmalloc_array() eventually calls kvmalloc_node() which
does not accept non GFP_KERNEL compatible flag like GFP_NOIO, then
kmalloc_node() is called indeed to allocate physically continuous
pages. When system memory is under heavy pressure, and the requesting
size is large, there is high probability that allocating continueous
pages will fail.
But simply using GFP_KERNEL flag to call kvmalloc_array() is also
progblematic. In the code path where scribble_alloc() is called, the
raid array is suspended, if kvmalloc_node() triggers memory reclaim I/Os
and such I/Os go back to the suspend raid array, deadlock will happen.
What is desired here is to allocate non-physically (a.k.a virtually)
continuous pages and avoid memory reclaim I/Os. Michal Hocko suggests
to use the mmealloc sceope APIs to restrict memory reclaim I/O in
allocating context, specifically to call memalloc_noio_save() when
suspend the raid array and to call memalloc_noio_restore() when
resume the raid array.
This patch adds the memalloc scope APIs in mddev_suspend() and
mddev_resume(), to restrict memory reclaim I/Os during the raid array
is suspended. The benifit of adding the memalloc scope API in the
unified entry point mddev_suspend()/mddev_resume() is, no matter which
md raid array type (personality), we are sure the deadlock by recursive
memory reclaim I/O won't happen on the suspending context.
Please notice that the memalloc scope APIs only take effect on the raid
array suspending context, if the memory allocation is from another new
created kthread after raid array suspended, the recursive memory reclaim
I/Os won't be restricted. The mddev_suspend()/mddev_resume() entries are
used for the critical section where the raid metadata is modifying,
creating a kthread to allocate memory inside the critical section is
queer and very probably being buggy.
Fixes: b330e6a49d ("md: convert to kvmalloc")
Suggested-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Song Liu <songliubraving@fb.com>
It is not not necessary to add a newline for them since they don't exceed
80 characters, and it is not intutive to distinguish ->hot_add_disk() from
hot_add_disk() too.
Signed-off-by: Guoqing Jiang <guoqing.jiang@cloud.ionos.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
Since rdev->kobj is removed asynchronously, it is possible that the
rdev->kobj still exists when try to add the rdev again after rdev
is removed. But this path md_ioctl (HOT_ADD_DISK) -> hot_add_disk
-> bind_rdev_to_array missed it.
Signed-off-by: Guoqing Jiang <guoqing.jiang@cloud.ionos.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
Since the purpose of call flush_workqueue in new_dev_store is to ensure
md_delayed_delete() has completed, so we should check rdev->del_work is
pending or not.
To suppress lockdep warning, we have to check mddev->del_work while
md_delayed_delete is attached to rdev->del_work, so it is not aligned
to the purpose of flush workquee. So a new workqueue is needed to avoid
the awkward situation, and introduce a new func flush_rdev_wq to flush
the new workqueue after check if there was pending work.
Also like new_dev_store, ADD_NEW_DISK ioctl has the same purpose to flush
workqueue while it holds bdev->bd_mutex, so make the same change applies
to the ioctl to avoid similar lock issue.
And md_delayed_delete actually wants to delete rdev, so rename the function
to rdev_delayed_delete.
Signed-off-by: Guoqing Jiang <guoqing.jiang@cloud.ionos.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
Coly reported possible circular locking dependencyi with LOCKDEP enabled,
quote the below info from the detailed report [1].
[ 1607.673903] Chain exists of:
[ 1607.673903] kn->count#256 --> (wq_completion)md_misc -->
(work_completion)(&rdev->del_work)
[ 1607.673903]
[ 1607.827946] Possible unsafe locking scenario:
[ 1607.827946]
[ 1607.898780] CPU0 CPU1
[ 1607.952980] ---- ----
[ 1608.007173] lock((work_completion)(&rdev->del_work));
[ 1608.069690] lock((wq_completion)md_misc);
[ 1608.149887] lock((work_completion)(&rdev->del_work));
[ 1608.242563] lock(kn->count#256);
[ 1608.283238]
[ 1608.283238] *** DEADLOCK ***
[ 1608.283238]
[ 1608.354078] 2 locks held by kworker/5:0/843:
[ 1608.405152] #0: ffff8889eecc9948 ((wq_completion)md_misc){+.+.}, at:
process_one_work+0x42b/0xb30
[ 1608.512399] #1: ffff888a1d3b7e10
((work_completion)(&rdev->del_work)){+.+.}, at: process_one_work+0x42b/0xb30
[ 1608.632130]
Since works (rdev->del_work and mddev->del_work) are queued in md_misc_wq,
then lockdep_map lock is held if either of them are running, then both of
them try to hold kernfs lock by call kobject_del. Then if new_dev_store
or array_state_store are triggered by write to the related sysfs node, so
the write operation gets kernfs lock, but need the lockdep_map because all
of them would trigger flush_workqueue(md_misc_wq) finally, then the same
lockdep_map lock is needed.
To suppress the lockdep warnning, we should flush the workqueue in case the
related work is pending. And several works are attached to md_misc_wq, so
we need to check which work should be checked:
1. for __md_stop_writes, the purpose of call flush workqueue is ensure sync
thread is started if it was starting, so check mddev->del_work is pending
or not since md_start_sync is attached to mddev->del_work.
2. __md_stop flushes md_misc_wq to ensure event_work is done, check the
event_work is enough. Assume raid_{ctr,dtr} -> md_stop -> __md_stop doesn't
need the kernfs lock.
3. both new_dev_store (holds kernfs lock) and ADD_NEW_DISK ioctl (holds the
bdev->bd_mutex) call flush_workqueue to ensure md_delayed_delete has
completed, this case will be handled in next patch.
4. md_open flushes workqueue to ensure the previous md is disappeared, but
it holds bdev->bd_mutex then try to flush workqueue, so it is better to
check mddev->del_work as well to avoid potential lock issue, this will be
done in another patch.
[1]: https://marc.info/?l=linux-raid&m=158518958031584&w=2
Cc: Coly Li <colyli@suse.de>
Reported-by: Coly Li <colyli@suse.de>
Signed-off-by: Guoqing Jiang <guoqing.jiang@cloud.ionos.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
The function sanitize_restored_xstate() sanitizes user xstates of an XSAVE
buffer by clearing bits not in the input 'xfeatures' from the buffer's
header->xfeatures, effectively resetting those features back to the init
state.
When supervisor xstates are introduced, it is necessary to make sure only
user xstates are sanitized. Ensure supervisor bits in header->xfeatures
stay set and supervisor states are not modified.
To make names clear, also:
- Rename the function to sanitize_restored_user_xstate().
- Rename input parameter 'xfeatures' to 'user_xfeatures'.
- In __fpu__restore_sig(), rename 'xfeatures' to 'user_xfeatures'.
Signed-off-by: Yu-cheng Yu <yu-cheng.yu@intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Dave Hansen <dave.hansen@linux.intel.com>
Link: https://lkml.kernel.org/r/20200512145444.15483-7-yu-cheng.yu@intel.com
Introduce essential bss information for multiple vifs hw_scan in
mt7615_mcu_uni_add_dev adding the interface
Signed-off-by: Sean Wang <sean.wang@mediatek.com>
Co-developed-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Differentiate hw scan channel time between passive and active scan
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Felix Fietkau <nbd@nbd.name>
