[ Upstream commit f5fe24ef17 ]
On the x86-64 architecture even a failing cmpxchg grants exclusive
access to the cacheline, making it preferable to retry the failed op
immediately instead of stalling with the pause instruction.
To illustrate the impact, below are benchmark results obtained by
running various will-it-scale tests on top of the 6.2-rc3 kernel and
Cascade Lake (2 sockets * 24 cores * 2 threads) CPU.
All results in ops/s. Note there is some variance in re-runs, but the
code is consistently faster when contention is present.
open3 ("Same file open/close"):
proc stock no-pause
1 805603 814942 (+%1)
2 1054980 1054781 (-0%)
8 1544802 1822858 (+18%)
24 1191064 2199665 (+84%)
48 851582 1469860 (+72%)
96 609481 1427170 (+134%)
fstat2 ("Same file fstat"):
proc stock no-pause
1 3013872 3047636 (+1%)
2 4284687 4400421 (+2%)
8 3257721 5530156 (+69%)
24 2239819 5466127 (+144%)
48 1701072 5256609 (+209%)
96 1269157 6649326 (+423%)
Additionally, a kernel with a private patch to help access() scalability:
access2 ("Same file access"):
proc stock patched patched
+nopause
24 2378041 2005501 5370335 (-15% / +125%)
That is, fixing the problems in access itself *reduces* scalability
after the cacheline ping-pong only happens in lockref with the pause
instruction.
Note that fstat and access benchmarks are not currently integrated into
will-it-scale, but interested parties can find them in pull requests to
said project.
Code at hand has a rather tortured history. First modification showed
up in commit d472d9d98b ("lockref: Relax in cmpxchg loop"), written
with Itanium in mind. Later it got patched up to use an arch-dependent
macro to stop doing it on s390 where it caused a significant regression.
Said macro had undergone revisions and was ultimately eliminated later,
going back to cpu_relax.
While I intended to only remove cpu_relax for x86-64, I got the
following comment from Linus:
I would actually prefer just removing it entirely and see if
somebody else hollers. You have the numbers to prove it hurts on
real hardware, and I don't think we have any numbers to the
contrary.
So I think it's better to trust the numbers and remove it as a
failure, than say "let's just remove it on x86-64 and leave
everybody else with the potentially broken code"
Additionally, Will Deacon (maintainer of the arm64 port, one of the
architectures previously benchmarked):
So, from the arm64 side of the fence, I'm perfectly happy just
removing the cpu_relax() calls from lockref.
As such, come back full circle in history and whack it altogether.
Signed-off-by: Mateusz Guzik <mjguzik@gmail.com>
Link: https://lore.kernel.org/all/CAGudoHHx0Nqg6DE70zAVA75eV-HXfWyhVMWZ-aSeOofkA_=WdA@mail.gmail.com/
Acked-by: Tony Luck <tony.luck@intel.com> # ia64
Acked-by: Nicholas Piggin <npiggin@gmail.com> # powerpc
Acked-by: Will Deacon <will@kernel.org> # arm64
Acked-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit be53771c87 ]
The Microsoft Devkit 2023 is a an ARM64 based machine featuring a
Realtek 8153 USB3.0-to-GBit Ethernet adapter. As in their other
machines, Microsoft uses a custom USB device ID.
Add the respective ID values to the driver. This makes Ethernet work on
the MS Devkit device. The chip has been visually confirmed to be a
RTL8153.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Link: https://lore.kernel.org/r/20230111133228.190801-1-andre.przywara@arm.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 8c6a42b5b0 ]
The SSI driver calls the AC'97 playback and transmit streams "AC97 Playback"
and "AC97 Capture" respectively. This is the same name used by the generic
AC'97 CODEC driver in ASoC, creating confusion for the Freescale ASoC card
when it attempts to use these widgets in routing. Add a "CPU" in the name
like the regular DAIs registered by the driver to disambiguate.
Acked-by: Shengjiu Wang <shengjiu.wang@gmail.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Link: https://lore.kernel.org/r/20230106-asoc-udoo-probe-v1-1-a5d7469d4f67@kernel.org
Signed-off-by: Mark Brown <broonie@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 08f0adb193 ]
Use NULL for NULL pointer to fix the following sparse warning:
drivers/cpufreq/armada-37xx-cpufreq.c:448:32: sparse: warning: Using plain integer as NULL pointer
Signed-off-by: Miles Chen <miles.chen@mediatek.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 45d619bdaf ]
Symbols _edata and _end in the linker script are the
only unaligned expicitly on page boundary. Although
_end is aligned implicitly by BSS_SECTION macro that
is still inconsistent and could lead to a bug if a tool
or function would assume that _edata is as aligned as
others.
For example, vmem_map_init() function does not align
symbols _etext, _einittext etc. Should these symbols
be unaligned as well, the size of ranges to update
were short on one page.
Instead of fixing every occurrence of this kind in the
code and external tools just force the alignment on
these two symbols.
Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 0d4d52361b ]
Using DEBUG_H without a prefix is very generic and inconsistent with
other header guards in arch/s390/include/asm. In fact it collides with
the same name in the ath9k wireless driver though that depends on !S390
via disabled wireless support. Let's just use a consistent header guard
name and prevent possible future trouble.
Signed-off-by: Niklas Schnelle <schnelle@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit eea8ce81fb ]
This modem has 7 interfaces, 5 of them are serial interfaces and are
driven by cdc_acm, while 2 of them are wwan interfaces and are driven
by cdc_ether:
If 0: Abstract (modem)
If 1: Abstract (modem)
If 2: Abstract (modem)
If 3: Abstract (modem)
If 4: Abstract (modem)
If 5: Ethernet Networking
If 6: Ethernet Networking
Without this change, the 2 network interfaces will be named to usb0
and usb1, our QA think the names are confusing and filed a bug on it.
After applying this change, the name will be wwan0 and wwan1, and
they could work well with modem manager.
Signed-off-by: Hui Wang <hui.wang@canonical.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230105034249.10433-1-hui.wang@canonical.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit faf28e240d ]
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>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 5b24ac2dfd ]
Size of the 'expect' array in the __report_matches is 1536 bytes, which
is exactly the default frame size warning limit of the xtensa
architecture.
As a result allmodconfig xtensa kernel builds with the gcc that does not
support the compiler plugins (which otherwise would push the said
warning limit to 2K) fail with the following message:
kernel/kcsan/kcsan_test.c:257:1: error: the frame size of 1680 bytes
is larger than 1536 bytes
Fix it by dynamically allocating the 'expect' array.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Reviewed-by: Marco Elver <elver@google.com>
Tested-by: Marco Elver <elver@google.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 01c5bb0cc2 ]
Tegra234 platform uses the tegra194-cpufreq driver, so add it
to the blocklist in cpufreq-dt-platdev driver to avoid the cpufreq
driver registration from there.
Signed-off-by: Sumit Gupta <sumitg@nvidia.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit a3be19b91e ]
It was observed that the kernel would potentially send
ISCSI_KEVENT_UNBIND_SESSION multiple times. Introduce 'target_state' in
iscsi_cls_session() to make sure session will send only one unbind session
event.
This introduces a regression wrt. the issue fixed in commit 13e60d3ba2
("scsi: iscsi: Report unbind session event when the target has been
removed"). If iscsid dies for any reason after sending an unbind session to
kernel, once iscsid is restarted, the kernel's ISCSI_KEVENT_UNBIND_SESSION
event is lost and userspace is then unable to logout. However, the session
is actually in invalid state (its target_id is INVALID) so iscsid should
not sync this session during restart.
Consequently we need to check the session's target state during iscsid
restart. If session is in unbound state, do not sync this session and
perform session teardown. This is OK because once a session is unbound, we
can not recover it any more (mainly because its target id is INVALID).
Signed-off-by: Wenchao Hao <haowenchao@huawei.com>
Link: https://lore.kernel.org/r/20221126010752.231917-1-haowenchao@huawei.com
Reviewed-by: Mike Christie <michael.christie@oracle.com>
Reviewed-by: Wu Bo <wubo40@huawei.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 300b655db1 ]
The initial default value of 0 for tp->rate_app_limited was incorrect,
since a flow is indeed application-limited until it first sends
data. Fixing the default to be 1 is generally correct but also
specifically will help user-space applications avoid using the initial
tcpi_delivery_rate value of 0 that persists until the connection has
some non-zero bandwidth sample.
Fixes: eb8329e0a0 ("tcp: export data delivery rate")
Suggested-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David Morley <morleyd@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Tested-by: David Morley <morleyd@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit fdfc76a116 ]
In the original implementation of dwmac5
commit 8bf993a587 ("net: stmmac: Add support for DWMAC5 and implement Safety Features")
all safety features were enabled by default.
Later it seems some implementations didn't have support for all the
features, so in
commit 5ac712dcdf ("net: stmmac: enable platform specific safety features")
the safety_feat_cfg structure was added to the callback and defined for
some platforms to selectively enable these safety features.
The problem is that only certain platforms were given that software
support. If the automotive safety package bit is set in the hardware
features register the safety feature callback is called for the platform,
and for platforms that didn't get a safety_feat_cfg defined this results
in the following NULL pointer dereference:
[ 7.933303] Call trace:
[ 7.935812] dwmac5_safety_feat_config+0x20/0x170 [stmmac]
[ 7.941455] __stmmac_open+0x16c/0x474 [stmmac]
[ 7.946117] stmmac_open+0x38/0x70 [stmmac]
[ 7.950414] __dev_open+0x100/0x1dc
[ 7.954006] __dev_change_flags+0x18c/0x204
[ 7.958297] dev_change_flags+0x24/0x6c
[ 7.962237] do_setlink+0x2b8/0xfa4
[ 7.965827] __rtnl_newlink+0x4ec/0x840
[ 7.969766] rtnl_newlink+0x50/0x80
[ 7.973353] rtnetlink_rcv_msg+0x12c/0x374
[ 7.977557] netlink_rcv_skb+0x5c/0x130
[ 7.981500] rtnetlink_rcv+0x18/0x2c
[ 7.985172] netlink_unicast+0x2e8/0x340
[ 7.989197] netlink_sendmsg+0x1a8/0x420
[ 7.993222] ____sys_sendmsg+0x218/0x280
[ 7.997249] ___sys_sendmsg+0xac/0x100
[ 8.001103] __sys_sendmsg+0x84/0xe0
[ 8.004776] __arm64_sys_sendmsg+0x24/0x30
[ 8.008983] invoke_syscall+0x48/0x114
[ 8.012840] el0_svc_common.constprop.0+0xcc/0xec
[ 8.017665] do_el0_svc+0x38/0xb0
[ 8.021071] el0_svc+0x2c/0x84
[ 8.024212] el0t_64_sync_handler+0xf4/0x120
[ 8.028598] el0t_64_sync+0x190/0x194
Go back to the original behavior, if the automotive safety package
is found to be supported in hardware enable all the features unless
safety_feat_cfg is passed in saying this particular platform only
supports a subset of the features.
Fixes: 5ac712dcdf ("net: stmmac: enable platform specific safety features")
Reported-by: Ning Cai <ncai@quicinc.com>
Signed-off-by: Andrew Halaney <ahalaney@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 6c54b7bc8a ]
put_device() shouldn't be called before a prior call to
device_register(). __thermal_cooling_device_register() doesn't follow
that properly and needs fixing. Also
thermal_cooling_device_destroy_sysfs() is getting called unnecessarily
on few error paths.
Fix all this by placing the calls at the right place.
Based on initial work done by Caleb Connolly.
Fixes: 4748f9687c ("thermal: core: fix some possible name leaks in error paths")
Fixes: c408b3d1d9 ("thermal: Validate new state in cur_state_store()")
Reported-by: Caleb Connolly <caleb.connolly@linaro.org>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Tested-by: Frank Rowand <frowand.list@gmail.com>
Reviewed-by: Yang Yingliang <yangyingliang@huawei.com>
Tested-by: Caleb Connolly <caleb.connolly@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit e49a1e1ee0 ]
Return an error pointer if ->get_max_state() fails. The current code
returns NULL which will cause an oops in the callers.
Fixes: c408b3d1d9 ("thermal: Validate new state in cur_state_store()")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Stable-dep-of: 6c54b7bc8a ("thermal: core: call put_device() only after device_register() fails")
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit c408b3d1d9 ]
In cur_state_store(), the new state of the cooling device is received
from user-space and is not validated by the thermal core but the same is
left for the individual drivers to take care of. Apart from duplicating
the code it leaves possibility for introducing bugs where a driver may
not do it right.
Lets make the thermal core check the new state itself and store the max
value in the cooling device structure.
Link: https://lore.kernel.org/all/Y0ltRJRjO7AkawvE@kili/
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Stable-dep-of: 6c54b7bc8a ("thermal: core: call put_device() only after device_register() fails")
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 6c977c5c2e ]
ALU table entry 2 register in KSZ9477 have bit positions reserved for
forwarding port map. This field is referred in ksz9477_fdb_del() for
clearing forward port map and alu table.
But current fdb_del refer ALU table entry 3 register for accessing forward
port map. Update ksz9477_fdb_del() to get forward port map from correct
alu table entry register.
With this bug, issue can be observed while deleting static MAC entries.
Delete any specific MAC entry using "bridge fdb del" command. This should
clear all the specified MAC entries. But it is observed that entries with
self static alone are retained.
Tested on LAN9370 EVB since ksz9477_fdb_del() is used common across
LAN937x and KSZ series.
Fixes: b987e98e50 ("dsa: add DSA switch driver for Microchip KSZ9477")
Signed-off-by: Rakesh Sankaranarayanan <rakesh.sankaranarayanan@microchip.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Link: https://lore.kernel.org/r/20230118174735.702377-1-rakesh.sankaranarayanan@microchip.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 903848249a ]
Avoid race between process wakeup and tpacket_v3 block timeout.
The test waits for cfg_timeout_msec for packets to arrive. Packets
arrive in tpacket_v3 rings, which pass packets ("frames") to the
process in batches ("blocks"). The sk waits for req3.tp_retire_blk_tov
msec to release a block.
Set the block timeout lower than the process waiting time, else
the process may find that no block has been released by the time it
scans the socket list. Convert to a ring of more than one, smaller,
blocks with shorter timeouts. Blocks must be page aligned, so >= 64KB.
Fixes: 5ebfb4cc30 ("selftests/net: toeplitz test")
Signed-off-by: Willem de Bruijn <willemb@google.com>
Link: https://lore.kernel.org/r/20230118151847.4124260-1-willemdebruijn.kernel@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 9be182da0a ]
In test_async_probe_init, second set of asynchronous devices are saved
in sync_dev[sync_id], which should be async_dev[async_id].
This makes these devices not unregistered when exit.
> modprobe test_async_driver_probe && \
> modprobe -r test_async_driver_probe && \
> modprobe test_async_driver_probe
...
> sysfs: cannot create duplicate filename '/devices/platform/test_async_driver.4'
> kobject_add_internal failed for test_async_driver.4 with -EEXIST,
don't try to register things with the same name in the same directory.
Fixes: 57ea974fb8 ("driver core: Rewrite test_async_driver_probe to cover serialization and NUMA affinity")
Signed-off-by: Chen Zhongjin <chenzhongjin@huawei.com>
Link: https://lore.kernel.org/r/20221125063541.241328-1-chenzhongjin@huawei.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 36225a7c72 ]
I got the following WARNING message while removing driver(ds2482):
------------[ cut here ]------------
do not call blocking ops when !TASK_RUNNING; state=1 set at [<000000002d50bfb6>] w1_process+0x9e/0x1d0 [wire]
WARNING: CPU: 0 PID: 262 at kernel/sched/core.c:9817 __might_sleep+0x98/0xa0
CPU: 0 PID: 262 Comm: w1_bus_master1 Tainted: G N 6.1.0-rc3+ #307
RIP: 0010:__might_sleep+0x98/0xa0
Call Trace:
exit_signals+0x6c/0x550
do_exit+0x2b4/0x17e0
kthread_exit+0x52/0x60
kthread+0x16d/0x1e0
ret_from_fork+0x1f/0x30
The state of task is set to TASK_INTERRUPTIBLE in loop in w1_process(),
set it to TASK_RUNNING when it breaks out of the loop to avoid the
warning.
Fixes: 3c52e4e627 ("W1: w1_process, block or sleep")
Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
Link: https://lore.kernel.org/r/20221205101558.3599162-1-yangyingliang@huawei.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 25d5648802 ]
I got a deadloop report while doing device(ds2482) add/remove test:
[ 162.241881] w1_master_driver w1_bus_master1: Waiting for w1_bus_master1 to become free: refcnt=1.
[ 163.272251] w1_master_driver w1_bus_master1: Waiting for w1_bus_master1 to become free: refcnt=1.
[ 164.296157] w1_master_driver w1_bus_master1: Waiting for w1_bus_master1 to become free: refcnt=1.
...
__w1_remove_master_device() can't return, because the dev->refcnt is not zero.
w1_add_master_device() |
w1_alloc_dev() |
atomic_set(&dev->refcnt, 2) |
kthread_run() |
|__w1_remove_master_device()
| kthread_stop()
// KTHREAD_SHOULD_STOP is set, |
// threadfn(w1_process) won't be |
// called. |
kthread() |
| // refcnt will never be 0, it's deadloop.
| while (atomic_read(&dev->refcnt)) {...}
After calling w1_add_master_device(), w1_process() is not really
invoked, before w1_process() starting, if kthread_stop() is called
in __w1_remove_master_device(), w1_process() will never be called,
the refcnt can not be decreased, then it causes deadloop in remove
function because of non-zero refcnt.
We need to make sure w1_process() is really started, so move the
set refcnt into w1_process() to fix this problem.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
Link: https://lore.kernel.org/r/20221205080434.3149205-1-yangyingliang@huawei.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 95e5fda3b5 ]
The interrupt handler (pt_core_irq_handler()) of the ptdma
driver can be called from interrupt context. The code flow
in this function can lead down to pt_core_execute_cmd() which
will attempt to grab a mutex, which is not appropriate in
interrupt context and ultimately leads to a kernel panic.
The fix here changes this mutex to a spinlock, which has
been verified to resolve the issue.
Fixes: fa5d823b16 ("dmaengine: ptdma: Initial driver for the AMD PTDMA")
Signed-off-by: Eric Pilmore <epilmore@gigaio.com>
Link: https://lore.kernel.org/r/20230119033907.35071-1-epilmore@gigaio.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 55ba18dc62 ]
The commit 4af1b64f80 ("octeontx2-pf: Fix lmtst ID used in aura
free") uses the get/put_cpu() to protect the usage of percpu pointer
in ->aura_freeptr() callback, but it also unnecessarily disable the
preemption for the blockable memory allocation. The commit 87b93b678e
("octeontx2-pf: Avoid use of GFP_KERNEL in atomic context") tried to
fix these sleep inside atomic warnings. But it only fix the one for
the non-rt kernel. For the rt kernel, we still get the similar warnings
like below.
BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:46
in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 1, name: swapper/0
preempt_count: 1, expected: 0
RCU nest depth: 0, expected: 0
3 locks held by swapper/0/1:
#0: ffff800009fc5fe8 (rtnl_mutex){+.+.}-{3:3}, at: rtnl_lock+0x24/0x30
#1: ffff000100c276c0 (&mbox->lock){+.+.}-{3:3}, at: otx2_init_hw_resources+0x8c/0x3a4
#2: ffffffbfef6537e0 (&cpu_rcache->lock){+.+.}-{2:2}, at: alloc_iova_fast+0x1ac/0x2ac
Preemption disabled at:
[<ffff800008b1908c>] otx2_rq_aura_pool_init+0x14c/0x284
CPU: 20 PID: 1 Comm: swapper/0 Tainted: G W 6.2.0-rc3-rt1-yocto-preempt-rt #1
Hardware name: Marvell OcteonTX CN96XX board (DT)
Call trace:
dump_backtrace.part.0+0xe8/0xf4
show_stack+0x20/0x30
dump_stack_lvl+0x9c/0xd8
dump_stack+0x18/0x34
__might_resched+0x188/0x224
rt_spin_lock+0x64/0x110
alloc_iova_fast+0x1ac/0x2ac
iommu_dma_alloc_iova+0xd4/0x110
__iommu_dma_map+0x80/0x144
iommu_dma_map_page+0xe8/0x260
dma_map_page_attrs+0xb4/0xc0
__otx2_alloc_rbuf+0x90/0x150
otx2_rq_aura_pool_init+0x1c8/0x284
otx2_init_hw_resources+0xe4/0x3a4
otx2_open+0xf0/0x610
__dev_open+0x104/0x224
__dev_change_flags+0x1e4/0x274
dev_change_flags+0x2c/0x7c
ic_open_devs+0x124/0x2f8
ip_auto_config+0x180/0x42c
do_one_initcall+0x90/0x4dc
do_basic_setup+0x10c/0x14c
kernel_init_freeable+0x10c/0x13c
kernel_init+0x2c/0x140
ret_from_fork+0x10/0x20
Of course, we can shuffle the get/put_cpu() to only wrap the invocation
of ->aura_freeptr() as what commit 87b93b678e does. But there are only
two ->aura_freeptr() callbacks, otx2_aura_freeptr() and
cn10k_aura_freeptr(). There is no usage of perpcu variable in the
otx2_aura_freeptr() at all, so the get/put_cpu() seems redundant to it.
We can move the get/put_cpu() into the corresponding callback which
really has the percpu variable usage and avoid the sprinkling of
get/put_cpu() in several places.
Fixes: 4af1b64f80 ("octeontx2-pf: Fix lmtst ID used in aura free")
Signed-off-by: Kevin Hao <haokexin@gmail.com>
Link: https://lore.kernel.org/r/20230118071300.3271125-1-haokexin@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 3f4ca5fafc ]
While one cpu is working on looking up the right socket from ehash
table, another cpu is done deleting the request socket and is about
to add (or is adding) the big socket from the table. It means that
we could miss both of them, even though it has little chance.
Let me draw a call trace map of the server side.
CPU 0 CPU 1
----- -----
tcp_v4_rcv() syn_recv_sock()
inet_ehash_insert()
-> sk_nulls_del_node_init_rcu(osk)
__inet_lookup_established()
-> __sk_nulls_add_node_rcu(sk, list)
Notice that the CPU 0 is receiving the data after the final ack
during 3-way shakehands and CPU 1 is still handling the final ack.
Why could this be a real problem?
This case is happening only when the final ack and the first data
receiving by different CPUs. Then the server receiving data with
ACK flag tries to search one proper established socket from ehash
table, but apparently it fails as my map shows above. After that,
the server fetches a listener socket and then sends a RST because
it finds a ACK flag in the skb (data), which obeys RST definition
in RFC 793.
Besides, Eric pointed out there's one more race condition where it
handles tw socket hashdance. Only by adding to the tail of the list
before deleting the old one can we avoid the race if the reader has
already begun the bucket traversal and it would possibly miss the head.
Many thanks to Eric for great help from beginning to end.
Fixes: 5e0724d027 ("tcp/dccp: fix hashdance race for passive sessions")
Suggested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jason Xing <kernelxing@tencent.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/lkml/20230112065336.41034-1-kerneljasonxing@gmail.com/
Link: https://lore.kernel.org/r/20230118015941.1313-1-kerneljasonxing@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 1c58420858 ]
Polling the completion can progress the request state to IDLE, either
inline with the completion, or through softirq. Either way, the state
may not be COMPLETED, so don't check for that. We only care if the state
isn't IN_FLIGHT.
This is fixing an issue where the driver aborts an IO that we just
completed. Seeing the "aborting" message instead of "polled" is very
misleading as to where the timeout problem resides.
Fixes: bf392a5dc0 ("nvme-pci: Remove tag from process cq")
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit e433adc60f ]
Currently, we run into a number of WARN()s when attempting to unload the
amdgpu driver (e.g. using "modprobe -r amdgpu"). These all stem from
calling drm_encoder_cleanup() too early. So, to fix this we can stop
calling drm_encoder_cleanup() from amdgpu_dm_fini() and instead have it
be called from amdgpu_dm_encoder_destroy(). Also, we don't need to free
in amdgpu_dm_encoder_destroy() since mst_encoders[] isn't explicitly
allocated by the slab allocator.
Fixes: f74367e492 ("drm/amdgpu/display: create fake mst encoders ahead of time (v4)")
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Hamza Mahfooz <hamza.mahfooz@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 9fb0db4051 ]
Avoid getting DFS referral from an exclusive lock in
cache_refresh_path() because the tcon IPC used for getting the
referral could be disconnected and thus causing a deadlock as shown
below:
task A task B
====== ======
cifs_demultiplex_thread() dfs_cache_find()
cifs_handle_standard() cache_refresh_path()
reconnect_dfs_server() down_write()
dfs_cache_noreq_find() get_dfs_referral()
down_read() <- deadlock smb2_get_dfs_refer()
SMB2_ioctl()
cifs_send_recv()
compound_send_recv()
wait_for_response()
where task A cannot wake up task B because it is blocked on
down_read() due to the exclusive lock held in cache_refresh_path() and
therefore not being able to make progress.
Fixes: c9f7110399 ("cifs: keep referral server sessions alive")
Reviewed-by: Aurélien Aptel <aurelien.aptel@gmail.com>
Signed-off-by: Paulo Alcantara (SUSE) <pc@cjr.nz>
Signed-off-by: Steve French <stfrench@microsoft.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 3782c0d6ed ]
betopff_init() only checks the total sum of the report counts for each
report field to be at least 4, but hid_betopff_play() expects 4 report
fields.
A device advertising an output report with one field and 4 report counts
would pass the check but crash the kernel with a NULL pointer dereference
in hid_betopff_play().
Fixes: 52cd7785f3 ("HID: betop: add drivers/hid/hid-betopff.c")
Signed-off-by: Pietro Borrello <borrello@diag.uniroma1.it>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit d71ebe8114 ]
Commit a7766ef18b33("virtio_net: disable cb aggressively") enables
virtqueue callback via the following statement:
do {
if (use_napi)
virtqueue_disable_cb(sq->vq);
free_old_xmit_skbs(sq, false);
} while (use_napi && kick &&
unlikely(!virtqueue_enable_cb_delayed(sq->vq)));
When NAPI is used and kick is false, the callback won't be enabled
here. And when the virtqueue is about to be full, the tx will be
disabled, but we still don't enable tx interrupt which will cause a TX
hang. This could be observed when using pktgen with burst enabled.
TO be consistent with the logic that tries to disable cb only for
NAPI, fixing this by trying to enable delayed callback only when NAPI
is enabled when the queue is about to be full.
Fixes: a7766ef18b ("virtio_net: disable cb aggressively")
Signed-off-by: Jason Wang <jasowang@redhat.com>
Tested-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 7b90f5a665 ]
PTP TX timestamp handling was observed to be broken with this driver
when using the raw Layer 2 PTP encapsulation. ptp4l was not receiving
the expected TX timestamp after transmitting a packet, causing it to
enter a failure state.
The problem appears to be due to the way that the driver pads packets
which are smaller than the Ethernet minimum of 60 bytes. If headroom
space was available in the SKB, this caused the driver to move the data
back to utilize it. However, this appears to cause other data references
in the SKB to become inconsistent. In particular, this caused the
ptp_one_step_sync function to later (in the TX completion path) falsely
detect the packet as a one-step SYNC packet, even when it was not, which
caused the TX timestamp to not be processed when it should be.
Using the headroom for this purpose seems like an unnecessary complexity
as this is not a hot path in the driver, and in most cases it appears
that there is sufficient tailroom to not require using the headroom
anyway. Remove this usage of headroom to prevent this inconsistency from
occurring and causing other problems.
Fixes: 653e92a917 ("net: macb: add support for padding and fcs computation")
Signed-off-by: Robert Hancock <robert.hancock@calian.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Claudiu Beznea <claudiu.beznea@microchip.com> # on SAMA7G5
Reviewed-by: Claudiu Beznea <claudiu.beznea@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 2c1e1b9490 ]
Eliminate anonymous module_init() and module_exit(), which can lead to
confusion or ambiguity when reading System.map, crashes/oops/bugs,
or an initcall_debug log.
Give each of these init and exit functions unique driver-specific
names to eliminate the anonymous names.
Example 1: (System.map)
ffffffff832fc78c t init
ffffffff832fc79e t init
ffffffff832fc8f8 t init
Example 2: (initcall_debug log)
calling init+0x0/0x12 @ 1
initcall init+0x0/0x12 returned 0 after 15 usecs
calling init+0x0/0x60 @ 1
initcall init+0x0/0x60 returned 0 after 2 usecs
calling init+0x0/0x9a @ 1
initcall init+0x0/0x9a returned 0 after 74 usecs
Fixes: e126ba97db ("mlx5: Add driver for Mellanox Connect-IB adapters")
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Eli Cohen <eli@mellanox.com>
Cc: Saeed Mahameed <saeedm@nvidia.com>
Cc: Leon Romanovsky <leon@kernel.org>
Cc: linux-rdma@vger.kernel.org
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit f51471d193 ]
According to HW spec element_type, element_attributes and parent_element_id fields
should be reserved (0x0) when calling MODIFY_SCHEDULING_ELEMENT command.
This patch remove initialization of these fields when calling the command.
Fixes: bd77bf1cb5 ("net/mlx5: Add SRIOV VF max rate configuration support")
Signed-off-by: Maor Dickman <maord@nvidia.com>
Reviewed-by: Eli Cohen <elic@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
