dc to pplib interface is changed for navi1x, renoir.
display_config_changed is not called by dc anymore.
smu_write_watermarks_table is not executed for navi1x, renoir
during boot up.
solution: call smu_write_watermarks_table just after dc pass
watermark clock settings to pplib
Signed-off-by: Hersen Wu <hersenxs.wu@amd.com>
Reviewed-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The timeout of identify cmd, which is invoked as part of admin queue
creation, can result in freeing of async event data both in
nvme_rdma_timeout handler and error handling path of
nvme_rdma_configure_admin queue thus causing NULL pointer reference.
Call Trace:
? nvme_rdma_setup_ctrl+0x223/0x800 [nvme_rdma]
nvme_rdma_create_ctrl+0x2ba/0x3f7 [nvme_rdma]
nvmf_dev_write+0xa54/0xcc6 [nvme_fabrics]
__vfs_write+0x1b/0x40
vfs_write+0xb2/0x1b0
ksys_write+0x61/0xd0
__x64_sys_write+0x1a/0x20
do_syscall_64+0x60/0x1e0
entry_SYSCALL_64_after_hwframe+0x44/0xa9
Reviewed-by: Roland Dreier <roland@purestorage.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Prabhath Sajeepa <psajeepa@purestorage.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
vtimes may wrap and time_before/after64() should be used to determine
whether a given vtime is before or after another. iocg_is_idle() was
incorrectly using plain "<" comparison do determine whether done_vtime
is before vtime. Here, the only thing we're interested in is whether
done_vtime matches vtime which indicates that there's nothing in
flight. Let's test for inequality instead.
Signed-off-by: Tejun Heo <tj@kernel.org>
Fixes: 7caa47151a ("blkcg: implement blk-iocost")
Cc: stable@vger.kernel.org # v5.4+
Signed-off-by: Jens Axboe <axboe@kernel.dk>
libtraceevent (used by perf and trace-cmd) failed to parse the
xhci_urb_dequeue trace event. This is because the user space trace
event format parsing is not a full C compiler. It can handle some basic
logic, but is not meant to be able to handle everything C can do.
In cases where a trace event field needs to be converted from a number
to a string, there's the __print_symbolic() macro that should be used:
See samples/trace_events/trace-events-sample.h
Some xhci trace events open coded the __print_symbolic() causing the
user spaces tools to fail to parse it. This has to be replaced with
__print_symbolic() instead.
CC: stable@vger.kernel.org
Reported-by: Tzvetomir Stoyanov <tstoyanov@vmware.com>
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=206531
Fixes: 5abdc2e6e1 ("usb: host: xhci: add urb_enqueue/dequeue/giveback tracers")
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Link: https://lore.kernel.org/r/20200306150858.21904-2-mathias.nyman@linux.intel.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Mika writes:
thunderbolt: Fix for v5.6-rc6
This includes a single commit that fixes incorrect return value from
tb_port_is_width_supported() if the read fails.
* tag 'thunderbolt-fix-for-v5.6-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/westeri/thunderbolt:
thunderbolt: Fix error code in tb_port_is_width_supported()
wq_select_unbound_cpu() is designed for unbound workqueues only, but
it's wrongly called when using a bound workqueue too.
Fixing this ensures work queued to a bound workqueue with
cpu=WORK_CPU_UNBOUND always runs on the local CPU.
Before, that would happen only if wq_unbound_cpumask happened to include
it (likely almost always the case), or was empty, or we got lucky with
forced round-robin placement. So restricting
/sys/devices/virtual/workqueue/cpumask to a small subset of a machine's
CPUs would cause some bound work items to run unexpectedly there.
Fixes: ef55718044 ("workqueue: schedule WORK_CPU_UNBOUND work on wq_unbound_cpumask CPUs")
Cc: stable@vger.kernel.org # v4.5+
Signed-off-by: Hillf Danton <hdanton@sina.com>
[dj: massage changelog]
Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Tejun Heo <tj@kernel.org>
If a GPIO we are trying to use is not available and we are deferring
the probe, don't output an error message.
This seems to have been the intent of commit 05c7477885
("i2c: gpio: Add support for named gpios in DT") but the error was
still output due to not checking the updated 'retdesc'.
Fixes: 05c7477885 ("i2c: gpio: Add support for named gpios in DT")
Signed-off-by: Hamish Martin <hamish.martin@alliedtelesis.co.nz>
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
Function i2c_dw_pci_remove() -> pci_free_irq_vectors() ->
pci_disable_msi() -> free_msi_irqs() will throw a BUG_ON() for MSI
enabled device since the driver has not released the requested IRQ before
calling the pci_free_irq_vectors().
Here driver requests an IRQ using devm_request_irq() but automatic
release happens only after remove callback. Fix this by explicitly
freeing the IRQ before calling pci_free_irq_vectors().
Fixes: 21aa3983d6 ("i2c: designware-pci: Switch over to MSI interrupts")
Cc: stable@vger.kernel.org # v5.4+
Signed-off-by: Jarkko Nikula <jarkko.nikula@linux.intel.com>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
Similar to the commit 02d715b4a8 ("iommu/vt-d: Fix RCU list debugging
warnings"), there are several other places that call
list_for_each_entry_rcu() outside of an RCU read side critical section
but with dmar_global_lock held. Silence those false positives as well.
drivers/iommu/intel-iommu.c:4288 RCU-list traversed in non-reader section!!
1 lock held by swapper/0/1:
#0: ffffffff935892c8 (dmar_global_lock){+.+.}, at: intel_iommu_init+0x1ad/0xb97
drivers/iommu/dmar.c:366 RCU-list traversed in non-reader section!!
1 lock held by swapper/0/1:
#0: ffffffff935892c8 (dmar_global_lock){+.+.}, at: intel_iommu_init+0x125/0xb97
drivers/iommu/intel-iommu.c:5057 RCU-list traversed in non-reader section!!
1 lock held by swapper/0/1:
#0: ffffffffa71892c8 (dmar_global_lock){++++}, at: intel_iommu_init+0x61a/0xb13
Signed-off-by: Qian Cai <cai@lca.pw>
Acked-by: Lu Baolu <baolu.lu@linux.intel.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
There are several places traverse RCU-list without holding any lock in
intel_iommu_init(). Fix them by acquiring dmar_global_lock.
WARNING: suspicious RCU usage
-----------------------------
drivers/iommu/intel-iommu.c:5216 RCU-list traversed in non-reader section!!
other info that might help us debug this:
rcu_scheduler_active = 2, debug_locks = 1
no locks held by swapper/0/1.
Call Trace:
dump_stack+0xa0/0xea
lockdep_rcu_suspicious+0x102/0x10b
intel_iommu_init+0x947/0xb13
pci_iommu_init+0x26/0x62
do_one_initcall+0xfe/0x500
kernel_init_freeable+0x45a/0x4f8
kernel_init+0x11/0x139
ret_from_fork+0x3a/0x50
DMAR: Intel(R) Virtualization Technology for Directed I/O
Fixes: d8190dc638 ("iommu/vt-d: Enable DMA remapping after rmrr mapped")
Signed-off-by: Qian Cai <cai@lca.pw>
Acked-by: Lu Baolu <baolu.lu@linux.intel.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Martin noticed that nct6775 driver does not load properly on his system
in v5.4+ kernels. The issue was bisected to commit b84398d6d7 ("i2c:
i801: Use iTCO version 6 in Cannon Lake PCH and beyond") but it is
likely not the culprit because the faulty code has been in the driver
already since commit 9424693035 ("i2c: i801: Create iTCO device on
newer Intel PCHs"). So more likely some commit that added PCI IDs of
recent chipsets made the driver to create the iTCO_wdt device on Martins
system.
The issue was debugged to be PCI configuration access to the PMC device
that is not present. This returns all 1's when read and this caused the
iTCO_wdt driver to accidentally request resourses used by nct6775.
It turns out that the SMI resource is only required for some ancient
systems, not the ones supported by this driver. For this reason do not
populate the SMI resource at all and drop all the related code. The
driver now always populates the main I/O resource and only in case of SPT
(Intel Sunrisepoint) compatible devices it adds another resource for the
NO_REBOOT bit. These two resources are of different types so
platform_get_resource() used by the iTCO_wdt driver continues to find
the both resources at index 0.
Link: https://lore.kernel.org/linux-hwmon/CAM1AHpQ4196tyD=HhBu-2donSsuogabkfP03v1YF26Q7_BgvgA@mail.gmail.com/
Fixes: 9424693035 ("i2c: i801: Create iTCO device on newer Intel PCHs")
[wsa: complete fix needs all of http://patchwork.ozlabs.org/project/linux-i2c/list/?series=160959&state=*]
Reported-by: Martin Volf <martin.volf.42@gmail.com>
Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Reviewed-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
The iTCO_wdt driver only needs ICH_RES_IO_SMI I/O resource when either
turn_SMI_watchdog_clear_off module parameter is set to match ->iTCO_version
(or higher), and when legacy iTCO_vendorsupport is set. Modify the driver
so that ICH_RES_IO_SMI is optional if the two conditions are not met.
Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Reviewed-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
In preparation for making ->smi_res optional the iTCO_wdt driver needs
to know whether vendorsupport is being set to non-zero. For this reason
export the variable.
Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Reviewed-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
Pull cpupower utility fix for v5.6 from Shuah Khan:
"This cpupower update for Linux 5.6-rc6 consists of a fix from
Mike Gilbert for build failures when -fno-common is enabled.
-fno-common will be default in gcc v10."
* tag 'linux-cpupower-5.6-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/shuah/linux:
cpupower: avoid multiple definition with gcc -fno-common
Simon Wunderlich says:
====================
Here is a batman-adv bugfix:
- Don't schedule OGM for disabled interface, by Sven Eckelmann
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
What the driver writes into MAC_MAXLEN_CFG does not actually represent
VLAN_ETH_FRAME_LEN but instead ETH_FRAME_LEN + ETH_FCS_LEN. Yes they are
numerically equal, but the difference is important, as the switch treats
VLAN-tagged traffic specially and knows to increase the maximum accepted
frame size automatically. So it is always wrong to account for VLAN in
the MAC_MAXLEN_CFG register.
Unconditionally increase the maximum allowed frame size for
double-tagged traffic. Accounting for the additional length does not
mean that the other VLAN membership checks aren't performed, so there's
no harm done.
Also, stop abusing the MTU name for configuring the MRU. There is no
support for configuring the MRU on an interface at the moment.
Fixes: a556c76adc ("net: mscc: Add initial Ocelot switch support")
Fixes: fa914e9c4d ("net: mscc: ocelot: create a helper for changing the port MTU")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit e18b353f10 ("ipvlan: add cond_resched_rcu() while
processing muticast backlog") added a cond_resched_rcu() in a loop
using rcu protection to iterate over slaves.
This is breaking rcu rules, so lets instead use cond_resched()
at a point we can reschedule
Fixes: e18b353f10 ("ipvlan: add cond_resched_rcu() while processing muticast backlog")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Mahesh Bandewar <maheshb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In our production environment we have faced with problem that updating
classid in cgroup with heavy tasks cause long freeze of the file tables
in this tasks. By heavy tasks we understand tasks with many threads and
opened sockets (e.g. balancers). This freeze leads to an increase number
of client timeouts.
This patch implements following logic to fix this issue:
аfter iterating 1000 file descriptors file table lock will be released
thus providing a time gap for socket creation/deletion.
Now update is non atomic and socket may be skipped using calls:
dup2(oldfd, newfd);
close(oldfd);
But this case is not typical. Moreover before this patch skip is possible
too by hiding socket fd in unix socket buffer.
New sockets will be allocated with updated classid because cgroup state
is updated before start of the file descriptors iteration.
So in common cases this patch has no side effects.
Signed-off-by: Dmitry Yakunin <zeil@yandex-team.ru>
Reviewed-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
The Rx bound multicast packets are deferred to a workqueue and
macvlan can also suffer from the same attack that was discovered
by Syzbot for IPvlan. This solution is not as effective as in
IPvlan. IPvlan defers all (Tx and Rx) multicast packet processing
to a workqueue while macvlan does this way only for the Rx. This
fix should address the Rx codition to certain extent.
Tx is still suseptible. Tx multicast processing happens when
.ndo_start_xmit is called, hence we cannot add cond_resched().
However, it's not that severe since the user which is generating
/ flooding will be affected the most.
Fixes: 412ca1550c ("macvlan: Move broadcasts into a work queue")
Signed-off-by: Mahesh Bandewar <maheshb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
IPvlan in L3 mode discards outbound multicast packets but performs
the check before ensuring the ether-header is set or not. This is
an error that Eric found through code browsing.
Fixes: 2ad7bf3638 (“ipvlan: Initial check-in of the IPVLAN driver.”)
Signed-off-by: Mahesh Bandewar <maheshb@google.com>
Reported-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It's a resource, not a parameter, so we can't copy it into the new
channel's TX queues, otherwise aliasing will lead to resource-
management bugs if the channel is subsequently torn down without
being initialised.
Before the Fixes:-tagged commit there was a similar bug with
tsoh_page, but I'm not sure it's worth doing another fix for such
old kernels.
Fixes: e9117e5099 ("sfc: Firmware-Assisted TSO version 2")
Suggested-by: Derek Shute <Derek.Shute@stratus.com>
Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull Ktest fixes and clean ups from Steven Rostedt:
- Make the default option oldconfig instead of randconfig (one too many
times I lost my config because I left the build type out)
- Add timeout to ssh sync to sync before reboot (prevents test hangs)
- A couple of spelling fix patches
* tag 'ktest-v5.6' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-ktest:
ktest: Fix typos in ktest.pl
ktest: Add timeout for ssh sync testing
ktest: Make default build option oldconfig not randconfig
ktest: Fix some typos in sample.conf
Pull MMC host fixes from Ulf Hansson:
- sdhci-msm: Silence warning about turning function into static
- sdhci-pci-gli: Fix support for GL975x by enabling MSI interrupt
* tag 'mmc-v5.6-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc:
mmc: sdhci-pci-gli: Enable MSI interrupt for GL975x
mmc: sdhci-msm: Mark sdhci_msm_cqe_disable static
Pull virtio fixes from Michael Tsirkin:
"Some bug fixes all over the place"
* tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost:
virtio_balloon: Adjust label in virtballoon_probe
virtio-blk: improve virtqueue error to BLK_STS
virtio-blk: fix hw_queue stopped on arbitrary error
virtio_ring: Fix mem leak with vring_new_virtqueue()
The alloc_pid() codepath used to be simpler. With the introducation of the
ability to choose specific pids in 49cb2fc42c ("fork: extend clone3() to
support setting a PID") it got more complex. It hasn't been super obvious
that ENOMEM is returned when the pid namespace init process/child subreaper
of the pid namespace has died. As can be seen from multiple attempts to
improve this see e.g. [1] and most recently [2].
We regressed returning ENOMEM in [3] and [2] restored it. Let's add a
comment on top explaining that this is historic and documented behavior and
cannot easily be changed.
[1]: 35f71bc0a0 ("fork: report pid reservation failure properly")
[2]: b26ebfe12f ("pid: Fix error return value in some cases")
[3]: 49cb2fc42c ("fork: extend clone3() to support setting a PID")
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
The recent futex inode life time fix changed the ordering of the futex key
union struct members, but forgot to adjust the hash function accordingly,
As a result the hashing omits the leading 64bit and even hashes beyond the
futex key causing a bad hash distribution which led to a ~100% performance
regression.
Hand in the futex key pointer instead of a random struct member and make
the size calculation based of the struct offset.
Fixes: 8019ad13ef ("futex: Fix inode life-time issue")
Reported-by: Rong Chen <rong.a.chen@intel.com>
Decoded-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Rong Chen <rong.a.chen@intel.com>
Link: https://lkml.kernel.org/r/87h7yy90ve.fsf@nanos.tec.linutronix.de
Before rebooting the box, a "ssh sync" is called to the test machine to see
if it is alive or not. But if the test machine is in a partial state, that
ssh may never actually finish, and the ktest test hangs.
Add a 10 second timeout to the sync test, which will fail after 10 seconds
and then cause the test to reboot the test machine.
Cc: stable@vger.kernel.org
Fixes: 6474ace999 ("ktest.pl: Powercycle the box on reboot if no connection can be made")
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
For the last time, I screwed up my ktest config file, and the build went
into the default "randconfig", blowing away the .config that I had set up.
The reason for the default randconfig was because when this was first
written, I wanted to do a bunch of randconfigs. But as time progressed,
ktest isn't about randconfig anymore, and because randconfig destroys the
config in the build directory, it's a dangerous default to have. Use
oldconfig as the default.
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
[why]
nv14 previously inherited soc bb from generic dcn 2, did not match
watermark values according to memory team
[how]
add nv14 specific soc bb: copy nv2 generic that it was
using from before, but changed num channels to 8
Signed-off-by: Martin Leung <martin.leung@amd.com>
Reviewed-by: Jun Lei <Jun.Lei@amd.com>
Acked-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The hierarchical parts of MSM pinctrl/GPIO is only
used when the device tree has a "wakeup-parent" as
a phandle, but the .irq_eoi is anyway assigned leading
to semantic problems on elder Qualcomm chipsets.
When the drivers/mfd/qcom-pm8xxx.c driver calls
chained_irq_exit() that call will in turn call chip->irq_eoi()
which is set to irq_chip_eoi_parent() by default on a
hierachical IRQ chip, and the parent is pinctrl-msm.c
so that will in turn unconditionally call
irq_chip_eoi_parent() again, but its parent is invalid
so we get the following crash:
Unnable to handle kernel NULL pointer dereference at
virtual address 00000010
pgd = (ptrval)
[00000010] *pgd=00000000
Internal error: Oops: 5 [#1] PREEMPT SMP ARM
(...)
PC is at irq_chip_eoi_parent+0x4/0x10
LR is at pm8xxx_irq_handler+0x1b4/0x2d8
If we solve this crash by avoiding to call up to
irq_chip_eoi_parent(), the machine will hang and get
reset by the watchdog, because of semantic issues,
probably inside irq_chip.
As a solution, just assign the .irq_eoi conditionally if
we are actually using a wakeup parent.
Cc: David Heidelberg <david@ixit.cz>
Cc: Bjorn Andersson <bjorn.andersson@linaro.org>
Cc: Lina Iyer <ilina@codeaurora.org>
Cc: Stephen Boyd <swboyd@chromium.org>
Cc: stable@vger.kernel.org
Fixes: e35a6ae0eb ("pinctrl/msm: Setup GPIO chip in hierarchy")
Link: https://lore.kernel.org/r/20200306121221.1231296-1-linus.walleij@linaro.org
Link: https://lore.kernel.org/r/20200309125207.571840-1-linus.walleij@linaro.org
Link: https://lore.kernel.org/r/20200309152604.585112-1-linus.walleij@linaro.org
Tested-by: David Heidelberg <david@ixit.cz>
Acked-by: Marc Zyngier <maz@kernel.org>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Each OSS PCM plugins allocate its internal buffer per pre-calculation
of the max buffer size through the chain of plugins (calling
src_frames and dst_frames callbacks). This works for most plugins,
but the rate plugin might behave incorrectly. The calculation in the
rate plugin involves with the fractional position, i.e. it may vary
depending on the input position. Since the buffer size
pre-calculation is always done with the offset zero, it may return a
shorter size than it might be; this may result in the out-of-bound
access as spotted by fuzzer.
This patch addresses those possible buffer overflow accesses by simply
setting the upper limit per the given buffer size for each plugin
before src_frames() and after dst_frames() calls.
Reported-by: syzbot+e1fe9f44fb8ecf4fb5dd@syzkaller.appspotmail.com
Cc: <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/000000000000b25ea005a02bcf21@google.com
Link: https://lore.kernel.org/r/20200309082148.19855-1-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
In commit 1ec17dbd90 ("inet_diag: fix reporting cgroup classid and
fallback to priority") croup classid reporting was fixed. But this works
only for TCP sockets because for other socket types icsk parameter can
be NULL and classid code path is skipped. This change moves classid
handling to inet_diag_msg_attrs_fill() function.
Also inet_diag_msg_attrs_size() helper was added and addends in
nlmsg_new() were reordered to save order from inet_sk_diag_fill().
Fixes: 1ec17dbd90 ("inet_diag: fix reporting cgroup classid and fallback to priority")
Signed-off-by: Dmitry Yakunin <zeil@yandex-team.ru>
Reviewed-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
ACS (auto PAD/FCS stripping) removes FCS off 802.3 packets (LLC) so that
there is no need to manually strip it for such packets. The enhanced DMA
descriptors allow to flag LLC packets so that the receiving callback can
use that to strip FCS manually or not. On the other hand, normal
descriptors do not support that.
Thus in order to not truncate LLC packet ACS should be disabled when
using normal DMA descriptors.
Fixes: 47dd7a540b ("net: add support for STMicroelectronics Ethernet controllers.")
Signed-off-by: Remi Pommarel <repk@triplefau.lt>
Signed-off-by: David S. Miller <davem@davemloft.net>
There is a problem when ipvlan slaves are created on a master device that
is a vmxnet3 device (ipvlan in VMware guests). The vmxnet3 driver does not
support unicast address filtering. When an ipvlan device is brought up in
ipvlan_open(), the ipvlan driver calls dev_uc_add() to add the hardware
address of the vmxnet3 master device to the unicast address list of the
master device, phy_dev->uc. This inevitably leads to the vmxnet3 master
device being forced into promiscuous mode by __dev_set_rx_mode().
Promiscuous mode is switched on the master despite the fact that there is
still only one hardware address that the master device should use for
filtering in order for the ipvlan device to be able to receive packets.
The comment above struct net_device describes the uc_promisc member as a
"counter, that indicates, that promiscuous mode has been enabled due to
the need to listen to additional unicast addresses in a device that does
not implement ndo_set_rx_mode()". Moreover, the design of ipvlan
guarantees that only the hardware address of a master device,
phy_dev->dev_addr, will be used to transmit and receive all packets from
its ipvlan slaves. Thus, the unicast address list of the master device
should not be modified by ipvlan_open() and ipvlan_stop() in order to make
ipvlan a workable option on masters that do not support unicast address
filtering.
Fixes: 2ad7bf3638 ("ipvlan: Initial check-in of the IPVLAN driver")
Reported-by: Per Sundstrom <per.sundstrom@redqube.se>
Signed-off-by: Jiri Wiesner <jwiesner@suse.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Mahesh Bandewar <maheshb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
After more careful studying, Paul informs me that we cannot rely on
ordering of RCU callbacks in the way that the the tagged commit did.
The current construct looks like this:
void C(struct rcu_head *rhp)
{
do_something(rhp);
call_rcu(&p->rh, B);
}
call_rcu(&p->rh, A);
call_rcu(&p->rh, C);
and we're relying on ordering between A and B, which isn't guaranteed.
Make this explicit instead, and have a work item issue the rcu_barrier()
to ensure that A has run before we manually execute B.
While thorough testing never showed this issue, it's dependent on the
per-cpu load in terms of RCU callbacks. The updated method simplifies
the code as well, and eliminates the need to maintain an rcu_head in
the fileset data.
Fixes: c1e2148f8e ("io_uring: free fixed_file_data after RCU grace period")
Reported-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
