The SG capability is inherently present with Multichannel DMA operation.
The register used to check for this capability with other DMA driver types
is not defined for MCDMA.
Several code sections incorrectly use struct xilinx_axidma_tx_segment
instead of struct xilinx_aximcdma_tx_segment when operating as
Multichannel DMA. As their structures are similar, this just works.
The driver maintains a list of shared memory buffers along with their
mapped buffer id's in a global linked list. These buffers need to be
unmapped after use by the user-space client.
The global shared memory list is initialized to zero entries in the
function amdtee_open(). This clearing of list entries can be a source
for memory leak on secure side if the global linked list previously
held some mapped buffer entries allocated from another TEE context.
Fix potential memory leak issue by moving global shared memory list
to AMD-TEE driver context data structure.
After merging the drm-misc tree, linux-next build (arm
multi_v7_defconfig) failed like this:
In file included from drivers/gpu/drm/nouveau/nouveau_ttm.c:26:
include/linux/swiotlb.h: In function 'swiotlb_max_mapping_size':
include/linux/swiotlb.h:99:9: error: 'SIZE_MAX' undeclared (first use in this function)
99 | return SIZE_MAX;
| ^~~~~~~~
include/linux/swiotlb.h:7:1: note: 'SIZE_MAX' is defined in header '<stdint.h>'; did you forget to '#include <stdint.h>'?
6 | #include <linux/init.h>
+++ |+#include <stdint.h>
7 | #include <linux/types.h>
include/linux/swiotlb.h:99:9: note: each undeclared identifier is reported only once for each function it appears in
99 | return SIZE_MAX;
| ^~~~~~~~
Due to bug in the bootloader, the PHY has floating address and may
randomly change on each PHY reset. To avoid it, the updated bootloader
with the following patch[0] should be used:
| ARM: protonic: disable on-die termination to fix PHY bootstrapping
|
| If on-die termination is enabled, the RXC pin of iMX6 will be pulled
| high. Since we already have an 10K pull-down on board, the RXC level on
| PHY reset will be ~800mV, which is mostly interpreted as 1. On some
| reboots we get 0 instead and kernel can't detect the PHY properly.
|
| Since the default 0x020e07ac value is 0, it is sufficient to remove this
| entry from the affected imxcfg files.
|
| Since we get stable 0 on pin PHYADDR[2], the PHY address is changed from
| 4 to 0.
With latest bootloader update, the PHY address will be fixed to "0".
The ZII devel B board has two generations of Marvell Switches. The
mv88e6352 supports an MDIO clock of 12MHz. However the older 88e6185
does not like 12MHz, and often fails to probe.
Reduce the clock speed to 5MHz, which seems to work reliably.
Cc: Chris Healy <cphealy@gmail.com> Fixes: b955387667ec ("ARM: dts: ZII: update MDIO speed and preamble") Signed-off-by: Andrew Lunn <andrew@lunn.ch> Reviewed-by: Chris Healy <cphealy@gmail.com> Signed-off-by: Shawn Guo <shawnguo@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
Commit bcf3440c6dd7 ("net: phy: micrel: add phy-mode support for the
KSZ9031 PHY") fixed micrel phy driver adding proper support for phy
modes. Adapt imx6q-udoo board phy settings : explicitly set required
delay configuration using "rgmii-id".
Fixes: cbd54fe0b2bc ("ARM: dts: imx6dl-udoo: Add board support based off imx6q-udoo") Signed-off-by: Sergey Matyukevich <geomatsi@gmail.com> Reviewed-by: Fabio Estevam <festevam@gmail.com> Signed-off-by: Shawn Guo <shawnguo@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
When streaming bluetooth audio, the sound is choppy due to the
fact that the default baud rate of the HCI interface is too slow
to handle 16-bit stereo at 48KHz.
The Bluetooth chip is capable of up to 4M baud on the serial port,
so this patch sets the max-speed to 4000000 in order to properly
stream audio over the Bluetooth.
Fixes: 593816fa2f35 ("arm64: dts: imx: Add Beacon i.MX8m-Mini development kit") Signed-off-by: Adam Ford <aford173@gmail.com> Signed-off-by: Shawn Guo <shawnguo@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
Add little-endian property to RCPM node (for ls1028a,ls1088a,ls208xa),
otherwise RCPM driver will program hardware with incorrect setting,
causing system (such as LS1028ARDB) failed to be waked by wakeup source.
Fixes: 791c88ca5713 (“arm64: dts: ls1028a: Add ftm_alarm0 DT node”) Fixes: f4fe3a8665495 (“arm64: dts: layerscape: add ftm_alarm0 node”) Signed-off-by: Biwen Li <biwen.li@nxp.com> Signed-off-by: Ran Wang <ran.wang_1@nxp.com> Acked-by: Li Yang <leoyang.li@nxp.com> Signed-off-by: Shawn Guo <shawnguo@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
DSA spec v1.1 [1] updated to include a stride size register for WQ
configuration that will specify how much space is reserved for the WQ
configuration register set. This change is expected to be in the final
gen1 DSA hardware. Fix the driver to use WQCFG_OFFSET() for all WQ
offset calculation and fixup WQCFG_OFFSET() to use the new calculated
wq size.
The Ethernet PHY on the Bananapi M64 has the RX and TX delays
enabled on the PHY, using pull-ups on the RXDLY and TXDLY pins.
Fix the phy-mode description to correct reflect this so that the
implementation doesn't reconfigure the delays incorrectly. This
happened with commit bbc4d71d6354 ("net: phy: realtek: fix rtl8211e
rx/tx delay config").
The Ethernet PHY on the Libre Computer ALL-H5-CC has the RX and TX
delays enabled on the PHY, using pull-ups on the RXDLY and TXDLY pins.
Fix the phy-mode description to correct reflect this so that the
implementation doesn't reconfigure the delays incorrectly. This
happened with commit bbc4d71d6354 ("net: phy: realtek: fix rtl8211e
rx/tx delay config").
The Ethernet PHY on the Bananapi M2+ has the RX and TX delays
enabled on the PHY, using pull-ups on the RXDLY and TXDLY pins.
Fix the phy-mode description to correct reflect this so that the
implementation doesn't reconfigure the delays incorrectly. This
happened with commit bbc4d71d6354 ("net: phy: realtek: fix rtl8211e
rx/tx delay config").
The Ethernet PHY on the Cubieboard 4 and A80 Optimus have the RX
and TX delays enabled on the PHY, using pull-ups on the RXDLY and
TXDLY pins.
Fix the phy-mode description to correct reflect this so that the
implementation doesn't reconfigure the delays incorrectly. This
happened with commit bbc4d71d6354 ("net: phy: realtek: fix rtl8211e
rx/tx delay config").
The Ethernet PHY on the Bananapi M3 and Cubietruck Plus have the RX
and TX delays enabled on the PHY, using pull-ups on the RXDLY and
TXDLY pins.
Fix the phy-mode description to correct reflect this so that the
implementation doesn't reconfigure the delays incorrectly. This
happened with commit bbc4d71d6354 ("net: phy: realtek: fix rtl8211e
rx/tx delay config").
Fixes: 039359948a4b ("ARM: dts: sun8i: a83t: Enable Ethernet on two boards") Signed-off-by: Chen-Yu Tsai <wens@csie.org> Signed-off-by: Maxime Ripard <maxime@cerno.tech> Acked-by: Jernej Skrabec <jernej.skrabec@siol.net> Link: https://lore.kernel.org/r/20201024162515.30032-6-wens@kernel.org Signed-off-by: Sasha Levin <sashal@kernel.org>
The Ethernet PHY on the Orange Pi Plus 2E has the RX and TX delays
enabled on the PHY, using pull-ups on the RXDLY and TXDLY pins.
Fix the phy-mode description to correct reflect this so that the
implementation doesn't reconfigure the delays incorrectly. This
happened with commit bbc4d71d6354 ("net: phy: realtek: fix rtl8211e
rx/tx delay config").
Fixes: 4904337fe34f ("ARM: dts: sunxi: Restore EMAC changes (boards)") Fixes: 7a78ef92cdc5 ("ARM: sun8i: h3: Enable EMAC with external PHY on Orange Pi Plus 2E") Signed-off-by: Chen-Yu Tsai <wens@csie.org> Signed-off-by: Maxime Ripard <maxime@cerno.tech> Tested-by: Jernej Skrabec <jernej.skrabec@siol.net> Acked-by: Jernej Skrabec <jernej.skrabec@siol.net> Link: https://lore.kernel.org/r/20201024162515.30032-5-wens@kernel.org Signed-off-by: Sasha Levin <sashal@kernel.org>
The Ethernet PHY on the Bananapi M1+ has the RX and TX delays
enabled on the PHY, using pull-ups on the RXDLY and TXDLY pins.
Fix the phy-mode description to correct reflect this so that the
implementation doesn't reconfigure the delays incorrectly. This
happened with commit bbc4d71d6354 ("net: phy: realtek: fix rtl8211e
rx/tx delay config").
The Ethernet PHY on the Cubietruck has the RX and TX delays
enabled on the PHY, using pull-ups on the RXDLY and TXDLY pins.
Fix the phy-mode description to correct reflect this so that the
implementation doesn't reconfigure the delays incorrectly. This
happened with commit bbc4d71d6354 ("net: phy: realtek: fix rtl8211e
rx/tx delay config").
The Ethernet PHY on the A31 Hummingbird has the RX and TX delays
enabled on the PHY, using pull-ups on the RXDLY and TXDLY pins.
Fix the phy-mode description to correct reflect this so that the
implementation doesn't reconfigure the delays incorrectly. This
happened with commit bbc4d71d6354 ("net: phy: realtek: fix rtl8211e
rx/tx delay config").
According to board schematic, PHY provides both, RX and TX delays.
However, according to "fix" Realtek provided for this board, only TX
delay should be provided by PHY.
Tests show that both variants work but TX only PHY delay works
slightly better.
Update ethernet node to reflect the fact that PHY provides TX delay.
Since commit bbc4d71d6354 ("net: phy: realtek: fix rtl8211e rx/tx delay config"),
the network is unusable on PineH64 model A.
This is due to phy-mode incorrectly set to rgmii instead of rgmii-id.
Fixes: 729e1ffcf47e ("arm64: allwinner: h6: add support for the Ethernet on Pine H64") Signed-off-by: Corentin Labbe <clabbe@baylibre.com> Signed-off-by: Maxime Ripard <maxime@cerno.tech> Link: https://lore.kernel.org/r/20201019063449.33316-1-clabbe@baylibre.com Signed-off-by: Sasha Levin <sashal@kernel.org>
Before the commit bbc4d71d6354 ("net: phy: realtek: fix rtl8211e rx/tx
delay config"), the software overwrite for RX/TX delays of the RTL8211e
were not working properly and the Beelink GS1 had both RX/TX delay of RGMII
interface set using pull-up on the TXDLY and RXDLY pins.
Now that these delays are working properly they overwrite the HW
config and set this to 'rgmii' meaning no delay on both RX/TX.
This makes the ethernet of this board not working anymore.
Set the phy-mode to 'rgmii-id' meaning RGMII with RX/TX delays
in the device-tree to keep the correct configuration.
The error_debugfs label is only used when either
CONFIG_USB_DWC2_PERIPHERAL or CONFIG_USB_DWC2_DUAL_ROLE is enabled. Add
the same #if to the error_debugfs label itself as the code which uses
this label already has.
This avoids the following compiler warning:
warning: label ‘error_debugfs’ defined but not used [-Wunused-label]
Fixes: e1c08cf23172ed ("usb: dwc2: Add missing cleanups when usb_add_gadget_udc() fails") Acked-by: Minas Harutyunyan <Minas.Harutyunyan@synopsys.com> Reported-by: kernel test robot <lkp@intel.com> Reported-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com> Signed-off-by: Felipe Balbi <balbi@kernel.org> Cc: stable@vger.kernel.org # 5.9.x Signed-off-by: Kamal Mostafa <kamal@canonical.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
QCOM KRYO2XX gold (big) silver (LITTLE) CPU cores are based on
Cortex-A73 and Cortex-A53 respectively and are meltdown safe,
hence add them to kpti_safe_list[].
Add MIDR value for KRYO2XX gold (big) and silver (LITTLE)
CPU cores which are used in Qualcomm Technologies, Inc.
SoCs. This will be used to identify and apply errata
which are applicable for these CPU cores.
Patch b2a846dbef4e ("gfs2: Ignore journal log writes for jdata holes")
tried (unsuccessfully) to fix a case in which writes were done to jdata
blocks, the blocks are sent to the ail list, then a punch_hole or truncate
operation caused the blocks to be freed. In other words, the ail items
are for jdata holes. Before b2a846dbef4e, the jdata hole caused function
gfs2_block_map to return -EIO, which was eventually interpreted as an
IO error to the journal, and then withdraw.
This patch changes function gfs2_get_block_noalloc, which is only used
for jdata writes, so it returns -ENODATA rather than -EIO, and when
-ENODATA is returned to gfs2_ail1_start_one, the error is ignored.
We can safely ignore it because gfs2_ail1_start_one is only called
when the jdata pages have already been written and truncated, so the
ail1 content no longer applies.
Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
To convert the number of pulses counted into an RPM estimation, we need
to divide by the width of our measurement interval instead of
multiplying by it. If the width of the measurement interval is zero we
don't update the RPM value to avoid dividing by zero.
We also don't need to do 64-bit division, with 32-bits we can handle a
fan running at over 4 million RPM.
__sb_start_write has some weird looking lockdep code that claims to
exist to handle nested freeze locking requests from xfs. The code as
written seems broken -- if we think we hold a read lock on any of the
higher freeze levels (e.g. we hold SB_FREEZE_WRITE and are trying to
lock SB_FREEZE_PAGEFAULT), it converts a blocking lock attempt into a
trylock.
However, it's not correct to downgrade a blocking lock attempt to a
trylock unless the downgrading code or the callers are prepared to deal
with that situation. Neither __sb_start_write nor its callers handle
this at all. For example:
sb_start_pagefault ignores the return value completely, with the result
that if xfs_filemap_fault loses a race with a different thread trying to
fsfreeze, it will proceed without pagefault freeze protection (thereby
breaking locking rules) and then unlocks the pagefault freeze lock that
it doesn't own on its way out (thereby corrupting the lock state), which
leads to a system hang shortly afterwards.
Normally, this won't happen because our ownership of a read lock on a
higher freeze protection level blocks fsfreeze from grabbing a write
lock on that higher level. *However*, if lockdep is offline,
lock_is_held_type unconditionally returns 1, which means that
percpu_rwsem_is_held returns 1, which means that __sb_start_write
unconditionally converts blocking freeze lock attempts into trylocks,
even when we *don't* hold anything that would block a fsfreeze.
Apparently this all held together until 5.10-rc1, when bugs in lockdep
caused lockdep to shut itself off early in an fstests run, and once
fstests gets to the "race writes with freezer" tests, kaboom. This
might explain the long trail of vanishingly infrequent livelocks in
fstests after lockdep goes offline that I've never been able to
diagnose.
We could fix it by spinning on the trylock if wait==true, but AFAICT the
locking works fine if lockdep is not built at all (and I didn't see any
complaints running fstests overnight), so remove this snippet entirely.
NOTE: Commit f4b554af9931 in 2015 created the current weird logic (which
used to exist in a different form in commit 5accdf82ba25c from 2012) in
__sb_start_write. XFS solved this whole problem in the late 2.6 era by
creating a variant of transactions (XFS_TRANS_NO_WRITECOUNT) that don't
grab intwrite freeze protection, thus making lockdep's solution
unnecessary. The commit claims that Dave Chinner explained that the
trylock hack + comment could be removed, but nobody ever did.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Jan Kara <jack@suse.cz> Signed-off-by: Sasha Levin <sashal@kernel.org>
Commit b2b29d6d0119 ("mm: account PMD tables like PTE tables") uncovered
a bug in uml, we forgot to call the destructor.
While we are here, give x a sane name.
Reported-by: Anton Ivanov <anton.ivanov@cambridgegreys.com> Co-developed-by: Matthew Wilcox (Oracle) <willy@infradead.org> Signed-off-by: Richard Weinberger <richard@nod.at> Tested-by: Christopher Obbard <chris.obbard@collabora.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Commit ce3d31ad3cac ("arm64/smp: Move rcu_cpu_starting() earlier") ensured
that RCU is informed early about incoming CPUs that might end up calling
into printk() before they are online. However, if such a CPU fails the
early CPU feature compatibility checks in check_local_cpu_capabilities(),
then it will be powered off or parked without informing RCU, leading to
an endless stream of stalls:
| rcu: INFO: rcu_preempt detected stalls on CPUs/tasks:
| rcu: 2-O...: (0 ticks this GP) idle=002/1/0x4000000000000000 softirq=0/0 fqs=2593
| (detected by 0, t=5252 jiffies, g=9317, q=136)
| Task dump for CPU 2:
| task:swapper/2 state:R running task stack: 0 pid: 0 ppid: 1 flags:0x00000028
| Call trace:
| ret_from_fork+0x0/0x30
Ensure that the dying CPU invokes rcu_report_dead() prior to being powered
off or parked.
cpu_psci_cpu_die() is called in the context of the dying CPU, which
will no longer be online or tracked by RCU. It is therefore not generally
safe to call printk() if the PSCI "cpu off" request fails, so remove the
pr_crit() invocation.
Cc: Qian Cai <cai@redhat.com> Cc: "Paul E. McKenney" <paulmck@kernel.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Link: https://lore.kernel.org/r/20201106103602.9849-2-will@kernel.org Signed-off-by: Will Deacon <will@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
In a surprising turn of events, it transpires that CPU capabilities
configured as ARM64_CPUCAP_WEAK_LOCAL_CPU_FEATURE are never set as the
result of late-onlining. Therefore our handling of erratum 1418040 does
not get activated if it is not required by any of the boot CPUs, even
though we allow late-onlining of an affected CPU.
In order to get things working again, replace the cpus_have_const_cap()
invocation with an explicit check for the current CPU using
this_cpu_has_cap().
Cc: Sai Prakash Ranjan <saiprakash.ranjan@codeaurora.org> Cc: Stephen Boyd <swboyd@chromium.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Mark Rutland <mark.rutland@arm.com> Reviewed-by: Suzuki K Poulose <suzuki.poulose@arm.com> Acked-by: Marc Zyngier <maz@kernel.org> Link: https://lore.kernel.org/r/20201106114952.10032-1-will@kernel.org Signed-off-by: Will Deacon <will@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
The Medion Akoya E2228T's ACPI _LID implementation is quite broken,
it has the same issues as the one from the Medion Akoya E2215T:
1. For notifications it uses an ActiveLow Edge GpioInt, rather then
an ActiveBoth one, meaning that the device is only notified when the
lid is closed, not when it is opened.
2. Matching with this its _LID method simply always returns 0 (closed)
In order for the Linux LID code to work properly with this implementation,
the lid_init_state selection needs to be set to ACPI_BUTTON_LID_INIT_OPEN,
add a DMI quirk for this.
While working on this I also found out that the MD60### part of the model
number differs per country/batch while all of the E2215T and E2228T models
have this issue, so also remove the " MD60198" part from the E2215T quirk.
Signed-off-by: Hans de Goede <hdegoede@redhat.com> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Fix the layout of 'struct desc64' to match the layout described in the
SDM Vol 3, Chapter 3 "Protected-Mode Memory Management", section 3.4.5
"Segment Descriptors", Figure 3-8 "Segment Descriptor". The test added
later in this series relies on this and crashes if this layout is not
correct.
Signed-off-by: Aaron Lewis <aaronlewis@google.com> Reviewed-by: Alexander Graf <graf@amazon.com>
Message-Id: <20201012194716.3950330-2-aaronlewis@google.com> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Use the uic_cmd->cmd_active as a flag to track the lifecycle of an UIC cmd.
The flag is set before sending the UIC cmd and cleared in IRQ handler. When
a PMC or UIC cmd completion timeout happens, if the flag is not set,
instead of returning timeout error, we still treat it as a successful
operation. This is to deal with the scenario in which completion has been
raised but the one waiting for the completion cannot be awaken in time due
to kernel scheduling problem.
The scsi_block_reqs_cnt increased in ufshcd_hold() is supposed to be
decreased back in ufshcd_ungate_work() in a paired way. However, if
specific ufshcd_hold/release sequences are met, it is possible that
scsi_block_reqs_cnt is increased twice but only one ungate work is
queued. To make sure scsi_block_reqs_cnt is handled by ufshcd_hold() and
ufshcd_ungate_work() in a paired way, increase it only if queue_work()
returns true.
Link: https://lore.kernel.org/r/1604384682-15837-2-git-send-email-cang@codeaurora.org Reviewed-by: Hongwu Su <hongwus@codeaurora.org> Reviewed-by: Stanley Chu <stanley.chu@mediatek.com> Reviewed-by: Bean Huo <beanhuo@micron.com> Signed-off-by: Can Guo <cang@codeaurora.org> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
The bpf_caps array is shorter without CAP_BPF, avoid out of bounds reads
if this isn't defined. Working around this avoids -Wno-array-bounds with
clang.
Turns out I was overly optimistic about cpu_pm blocking idle being a
solution for handling edge interrupts. While it helps in preventing
entering idle states that potentially lose context, we can still get
an edge interrupt triggering while entering idle. So we need to also
add back the workaround for seeing if there are any pending edge
interrupts when waking up.
Signed-off-by: Tony Lindgren <tony@atomide.com> Cc: Aaro Koskinen <aaro.koskinen@iki.fi> Cc: Grygorii Strashko <grygorii.strashko@ti.com> Cc: Keerthy <j-keerthy@ti.com> Cc: Ladislav Michl <ladis@linux-mips.org> Cc: Peter Ujfalusi <peter.ujfalusi@ti.com> Cc: Russell King <rmk+kernel@armlinux.org.uk> Cc: Tero Kristo <t-kristo@ti.com> Link: https://lore.kernel.org/r/20201028060556.56038-1-tony@atomide.com Signed-off-by: Linus Walleij <linus.walleij@linaro.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
In async_resync mode, we log the TCP seq of records until the async request
is completed. Later, in case one of the logged seqs matches the resync
request, we return it, together with its record serial number. Before this
fix, we mistakenly returned the serial number of the current record
instead.
Recycle the page running page_pool_put_full_page() in
mvneta_swbm_add_rx_fragment routine when the last descriptor
contains just the FCS or if the received packet contains more than
MAX_SKB_FRAGS fragments
Firmware is unable to retain the port counters during any kind of
fatal or non-fatal resets, so we must clear the port counters to
avoid false detection of port counter overflow.
Fixes: fea6b3335527 ("bnxt_en: Accumulate all counters.") Reviewed-by: Edwin Peer <edwin.peer@broadcom.com> Reviewed-by: Vasundhara Volam <vasundhara-v.volam@broadcom.com> Signed-off-by: Michael Chan <michael.chan@broadcom.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
bnxt_add_one_ctr() adds a hardware counter to a software counter and
adjusts for the hardware counter wraparound against the mask. The logic
assumes that the hardware counter is always smaller than or equal to
the mask.
This assumption is mostly correct. But in some cases if the firmware
is older and does not provide the accurate mask, the driver can use
a mask that is smaller than the actual hardware mask. This can cause
some extra carry bits to be added to the software counter, resulting in
counters that far exceed the actual value. Fix it by masking the
hardware counter with the mask passed into bnxt_add_one_ctr().
Fixes: fea6b3335527 ("bnxt_en: Accumulate all counters.") Reviewed-by: Vasundhara Volam <vasundhara-v.volam@broadcom.com> Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com> Reviewed-by: Edwin Peer <edwin.peer@broadcom.com> Signed-off-by: Michael Chan <michael.chan@broadcom.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
pm_runtime_get_sync() will increment pm usage at first and it will
resume the device later. If runtime of the device has error or
device is in inaccessible state(or other error state), resume
operation will fail. If we do not call put operation to decrease
the reference, it will result in reference count leak. Moreover,
this device cannot enter the idle state and always stay busy or other
non-idle state later. So we fixed it by replacing it with
pm_runtime_resume_and_get.
Fixes: 8fff755e9f8d0 ("net: fec: Ensure clocks are enabled while using mdio bus") Signed-off-by: Zhang Qilong <zhangqilong3@huawei.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In many case, we need to check return value of pm_runtime_get_sync, but
it brings a trouble to the usage counter processing. Many callers forget
to decrease the usage counter when it failed, which could resulted in
reference leak. It has been discussed a lot[0][1]. So we add a function
to deal with the usage counter for better coding.
[0]https://lkml.org/lkml/2020/6/14/88
[1]https://patchwork.ozlabs.org/project/linux-tegra/list/?series=178139 Signed-off-by: Zhang Qilong <zhangqilong3@huawei.com> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Currently when QoS is enabled for VF and any min_rate is configured,
the driver sets bw_share value to at least 1 and doesn’t allow to set
it to 0 to make minimal rate unlimited. It means there is always a
minimal rate configured for every VF, even if user tries to remove it.
In order to make QoS disable possible, check whether all vports have
configured min_rate = 0. If this is true, set their bw_share to 0 to
disable min_rate limitations.
Fixes: c9497c98901c ("net/mlx5: Add support for setting VF min rate") Signed-off-by: Vladyslav Tarasiuk <vladyslavt@nvidia.com> Reviewed-by: Moshe Shemesh <moshe@nvidia.com> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Currently, if user disables VFs with some min and max rates configured,
they are cleared. But QoS data is not cleared and restored upon next VF
enable placing limits on minimal rate for given VF, when user expects
none.
To match cleared vport->info struct with QoS-related min and max rates
upon VF disable, clear vport->qos struct too.
Fixes: a7ee18bdee83 ("RDMA/mlx5: Allow creating a matcher for a NIC TX flow table") Signed-off-by: Michael Guralnik <michaelgur@nvidia.com> Reviewed-by: Mark Bloch <mbloch@nvidia.com> Reviewed-by: Maor Gottlieb <maorg@nvidia.com> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Bond events handler uses bond_slave_get_rtnl to check if net device
is bond slave. bond_slave_get_rtnl return the rcu rx_handler pointer
from the netdev which exists for bond slaves but also exists for
devices that are attached to linux bridge so using it as indication
for bond slave is wrong.
Before commit c0cfa2d8a788 ("vsock: add multi-transports support"),
if a G2H transport was loaded (e.g. virtio transport), every packets
was forwarded to the host, regardless of the destination CID.
The H2G transports implemented until then (vhost-vsock, VMCI) always
responded with an error, if the destination CID was not
VMADDR_CID_HOST.
>From that commit, we are using the remote CID to decide which
transport to use, so packets with remote CID > VMADDR_CID_HOST(2)
are sent only through H2G transport. If no H2G is available, packets
are discarded directly in the guest.
Some use cases (e.g. Nitro Enclaves [1]) rely on the old behaviour
to implement sibling VMs communication, so we restore the old
behavior when no H2G is registered.
It will be up to the host to discard packets if the destination is
not the right one. As it was already implemented before adding
multi-transport support.
Tested with nested QEMU/KVM by me and Nitro Enclaves by Andra.
[1] Documentation/virt/ne_overview.rst
Cc: Jorgen Hansen <jhansen@vmware.com> Cc: Dexuan Cui <decui@microsoft.com> Fixes: c0cfa2d8a788 ("vsock: add multi-transports support") Reported-by: Andra Paraschiv <andraprs@amazon.com> Tested-by: Andra Paraschiv <andraprs@amazon.com> Signed-off-by: Stefano Garzarella <sgarzare@redhat.com> Link: https://lore.kernel.org/r/20201112133837.34183-1-sgarzare@redhat.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
During loss recovery, retransmitted packets are forced to use TCP
timestamps to calculate the RTT samples, which have a millisecond
granularity. BBR is designed using a microsecond granularity. As a
result, multiple RTT samples could be truncated to the same RTT value
during loss recovery. This is problematic, as BBR will not enter
PROBE_RTT if the RTT sample is <= the current min_rtt sample, meaning
that if there are persistent losses, PROBE_RTT will constantly be
pushed off and potentially never re-entered. This patch makes sure
that BBR enters PROBE_RTT by checking if RTT sample is < the current
min_rtt sample, rather than <=.
The Netflix transport/TCP team discovered this bug in the Linux TCP
BBR code during lab tests.
This is because it holds asoc when transport->proto_unreach_timer starts
and puts asoc when the timer stops, and without holding transport the
transport could be freed when the timer is still running.
So fix it by holding/putting transport instead for proto_unreach_timer
in transport, just like other timers in transport.
v1->v2:
- Also use sctp_transport_put() for the "out_unlock:" path in
sctp_generate_proto_unreach_event(), as Marcelo noticed.
Fixes: 50b5d6ad6382 ("sctp: Fix a race between ICMP protocol unreachable and connect()") Reported-by: Hangbin Liu <liuhangbin@gmail.com> Signed-off-by: Xin Long <lucien.xin@gmail.com> Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com> Link: https://lore.kernel.org/r/102788809b554958b13b95d33440f5448113b8d6.1605331373.git.lucien.xin@gmail.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The code refactoring of ILT configuration was not complete, the old
unused variables were used for the SRC block. That could lead to the memory
corruption by HW when rx filters are configured.
This patch completes that refactoring.
Fixes: 8a52bbab39c9 (qed: Debug feature: ilt and mdump) Signed-off-by: Igor Russkikh <irusskikh@marvell.com> Signed-off-by: Ariel Elior <aelior@marvell.com> Signed-off-by: Dmitry Bogdanov <dbogdanov@marvell.com> Link: https://lore.kernel.org/r/20201116132944.2055-1-dbogdanov@marvell.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Fix to return a negative error code from the error handling
case instead of 0, as done elsewhere in this function.
Fixes: 469981b17a4f ("qed: Add unaligned and packed packet processing") Fixes: fcb39f6c10b2 ("qed: Add mpa buffer descriptors for storing and processing mpa fpdus") Fixes: 1e28eaad07ea ("qed: Add iWARP support for fpdu spanned over more than two tcp packets") Reported-by: Hulk Robot <hulkci@huawei.com> Signed-off-by: Zhang Changzhong <zhangchangzhong@huawei.com> Acked-by: Michal Kalderon <michal.kalderon@marvell.com> Link: https://lore.kernel.org/r/1605532033-27373-1-git-send-email-zhangchangzhong@huawei.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The ethernet driver may allocate skb (and skb->data) via napi_alloc_skb().
This ends up to page_frag_alloc() to allocate skb->data from
page_frag_cache->va.
During the memory pressure, page_frag_cache->va may be allocated as
pfmemalloc page. As a result, the skb->pfmemalloc is always true as
skb->data is from page_frag_cache->va. The skb will be dropped if the
sock (receiver) does not have SOCK_MEMALLOC. This is expected behaviour
under memory pressure.
However, once kernel is not under memory pressure any longer (suppose large
amount of memory pages are just reclaimed), the page_frag_alloc() may still
re-use the prior pfmemalloc page_frag_cache->va to allocate skb->data. As a
result, the skb->pfmemalloc is always true unless page_frag_cache->va is
re-allocated, even if the kernel is not under memory pressure any longer.
Here is how kernel runs into issue.
1. The kernel is under memory pressure and allocation of
PAGE_FRAG_CACHE_MAX_ORDER in __page_frag_cache_refill() will fail. Instead,
the pfmemalloc page is allocated for page_frag_cache->va.
2: All skb->data from page_frag_cache->va (pfmemalloc) will have
skb->pfmemalloc=true. The skb will always be dropped by sock without
SOCK_MEMALLOC. This is an expected behaviour.
3. Suppose a large amount of pages are reclaimed and kernel is not under
memory pressure any longer. We expect skb->pfmemalloc drop will not happen.
4. Unfortunately, page_frag_alloc() does not proactively re-allocate
page_frag_alloc->va and will always re-use the prior pfmemalloc page. The
skb->pfmemalloc is always true even kernel is not under memory pressure any
longer.
Fix this by freeing and re-allocating the page instead of recycling it.
References: https://lore.kernel.org/lkml/20201103193239.1807-1-dongli.zhang@oracle.com/
References: https://lore.kernel.org/linux-mm/20201105042140.5253-1-willy@infradead.org/ Suggested-by: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Aruna Ramakrishna <aruna.ramakrishna@oracle.com> Cc: Bert Barbe <bert.barbe@oracle.com> Cc: Rama Nichanamatlu <rama.nichanamatlu@oracle.com> Cc: Venkat Venkatsubra <venkat.x.venkatsubra@oracle.com> Cc: Manjunath Patil <manjunath.b.patil@oracle.com> Cc: Joe Jin <joe.jin@oracle.com> Cc: SRINIVAS <srinivas.eeda@oracle.com> Fixes: 79930f5892e1 ("net: do not deplete pfmemalloc reserve") Signed-off-by: Dongli Zhang <dongli.zhang@oracle.com> Acked-by: Vlastimil Babka <vbabka@suse.cz> Reviewed-by: Eric Dumazet <edumazet@google.com> Link: https://lore.kernel.org/r/20201115201029.11903-1-dongli.zhang@oracle.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The x25_disconnect function in x25_subr.c would decrease the refcount of
"x25->neighbour" (struct x25_neigh) and reset this pointer to NULL.
However, the x25_rx_call_request function in af_x25.c, which is called
when we receive a connection request, does not increase the refcount when
it assigns the pointer.
Fix this issue by increasing the refcount of "struct x25_neigh" in
x25_rx_call_request.
This patch fixes frequent kernel crashes when using AF_X25 sockets.
Fixes: 4becb7ee5b3d ("net/x25: Fix x25_neigh refcnt leak when x25 disconnect") Cc: Martin Schiller <ms@dev.tdt.de> Signed-off-by: Xie He <xie.he.0141@gmail.com> Link: https://lore.kernel.org/r/20201112103506.5875-1-xie.he.0141@gmail.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If tcp socket has more data than Encrypted Handshake Message then
tls_sw_recvmsg will try to decrypt next record instead of returning
full control message to userspace as mentioned in comment. The next
message - usually Application Data - gets corrupted because it uses
zero copy for decryption that's why the data is not stored in skb
for next iteration. Revert check to not decrypt next record if
current is not Application Data.
Fixes: 692d7b5d1f91 ("tls: Fix recvmsg() to be able to peek across multiple records") Signed-off-by: Vadim Fedorenko <vfedorenko@novek.ru> Link: https://lore.kernel.org/r/1605413760-21153-1-git-send-email-vfedorenko@novek.ru Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
During rmnet unregistration, the real device rx_handler is first cleared
followed by the removal of rx_handler_data after the rcu synchronization.
Any packets in the receive path may observe that the rx_handler is NULL.
However, there is no check when dereferencing this value to use the
rmnet_port information.
This fixes following splat by adding the NULL check.
Unable to handle kernel NULL pointer dereference at virtual
address 000000000000000d
pc : rmnet_rx_handler+0x124/0x284
lr : rmnet_rx_handler+0x124/0x284
rmnet_rx_handler+0x124/0x284
__netif_receive_skb_core+0x758/0xd74
__netif_receive_skb+0x50/0x17c
process_backlog+0x15c/0x1b8
napi_poll+0x88/0x284
net_rx_action+0xbc/0x23c
__do_softirq+0x20c/0x48c
If a user unbinds and re-binds a NC-SI aware driver the kernel will
attempt to register the netlink interface at runtime. The structure is
marked __ro_after_init so registration fails spectacularly at this point.
Jakub pointed out that ncsi_register_dev is obviously broken, because
there is only one family so it would never work if there was more than
one ncsi netdev.
Fix the crash by registering the netlink family once on boot, and drop
the code to unregister it.
Fixes: 955dc68cb9b2 ("net/ncsi: Add generic netlink family") Signed-off-by: Joel Stanley <joel@jms.id.au> Reviewed-by: Samuel Mendoza-Jonas <sam@mendozajonas.com> Link: https://lore.kernel.org/r/20201112061210.914621-1-joel@jms.id.au Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
On resync, the driver calls inet_lookup_established
(__inet6_lookup_established) that increases sk_refcnt of the socket. To
decrease it, the driver set skb->destructor to sock_edemux. However, it
didn't work well, because the TCP stack also sets this destructor for
early demux, and the refcount gets decreased only once, while increased
two times (in mlx5e and in the TCP stack). It leads to a socket leak, a
TLS context leak, which in the end leads to calling tls_dev_del twice:
on socket close and on driver unload, which in turn leads to a crash.
This commit fixes the refcount leak by calling sock_gen_put right away
after using the socket, thus fixing all the subsequent issues.
Slave function read the following capabilities from the wrong offset:
1. log_mc_entry_sz
2. fs_log_entry_sz
3. log_mc_hash_sz
Fix that by adjusting these capabilities offset to match firmware
layout.
Due to the wrong offset read, the following issues might occur:
1+2. Negative value reported at max_mcast_qp_attach.
3. Driver to init FW with multicast hash size of zero.
Fixes: a40ded604365 ("net/mlx4_core: Add masking for a few queries on HCA caps") Signed-off-by: Aya Levin <ayal@nvidia.com> Reviewed-by: Moshe Shemesh <moshe@nvidia.com> Reviewed-by: Eran Ben Elisha <eranbe@nvidia.com> Signed-off-by: Tariq Toukan <tariqt@nvidia.com> Link: https://lore.kernel.org/r/20201118081922.553-1-tariqt@nvidia.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
A user reports (slightly shortened from the original message):
libphy: lantiq,xrx200-mdio: probed
mdio_bus 1e108000.switch-mii: MDIO device at address 17 is missing.
gswip 1e108000.switch lan: no phy at 2
gswip 1e108000.switch lan: failed to connect to port 2: -19
lantiq,xrx200-net 1e10b308.eth eth0: error -19 setting up slave phy
This is a single-port board using the internal Fast Ethernet PHY. The
user reports that switching to PHY scanning instead of configuring the
PHY within device-tree works around this issue.
The documentation for the standalone variant of the PHY11G (which is
probably very similar to what is used inside the xRX200 SoCs but having
the firmware burnt onto that standalone chip in the factory) states that
the PHY needs 300ms to be ready for MDIO communication after releasing
the reset.
Add a 300ms delay after initializing all GPHYs to ensure that the GPHY
firmware had enough time to initialize and to appear on the MDIO bus.
Unfortunately there is no (known) documentation on what the minimum time
to wait after releasing the reset on an internal PHY so play safe and
take the one for the external variant. Only wait after the last GPHY
firmware is loaded to not slow down the initialization too much (
xRX200 has two GPHYs but newer SoCs have at least three GPHYs).
Fixes: 14fceff4771e51 ("net: dsa: Add Lantiq / Intel DSA driver for vrx200") Reviewed-by: Andrew Lunn <andrew@lunn.ch> Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com> Acked-by: Hauke Mehrtens <hauke@hauke-m.de> Reviewed-by: Florian Fainelli <f.fainelli@gmail.com> Link: https://lore.kernel.org/r/20201115165757.552641-1-martin.blumenstingl@googlemail.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The current NetLabel code doesn't correctly keep track of the netlink
dump state in some cases, in particular when multiple interfaces with
large configurations are loaded. The problem manifests itself by not
reporting the full configuration to userspace, even though it is
loaded and active in the kernel. This patch fixes this by ensuring
that the dump state is properly reset when necessary inside the
netlbl_unlabel_staticlist() function.
Fixes: 8cc44579d1bd ("NetLabel: Introduce static network labels for unlabeled connections") Signed-off-by: Paul Moore <paul@paul-moore.com> Link: https://lore.kernel.org/r/160484450633.3752.16512718263560813473.stgit@sifl Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Transactions sit on one of several lists, depending on their state
(allocated, pending, complete, or polled). A spinlock protects
against concurrent access when transactions are moved between these
lists.
Transactions are also reference counted. A newly-allocated
transaction has an initial count of 1; a transaction is released in
gsi_trans_free() only if its decremented reference count reaches 0.
Releasing a transaction includes removing it from the polled (or if
unused, allocated) list, so the spinlock is acquired when we release
a transaction.
The reference count is used to allow a caller to synchronously wait
for a committed transaction to complete. In this case, the waiter
takes an extra reference to the transaction *before* committing it
(so it won't be freed), and releases its reference (calls
gsi_trans_free()) when it is done with it.
Similarly, gsi_channel_update() takes an extra reference to ensure a
transaction isn't released before the function is done operating on
it. Until the transaction is moved to the completed list (by this
function) it won't be freed, so this reference is taken "safely."
But in the quiesce path, we want to wait for the "last" transaction,
which we find in the completed or polled list. Transactions on
these lists can be freed at any time, so we (try to) prevent that
by taking the reference while holding the spinlock.
Currently gsi_trans_free() decrements a transaction's reference
count unconditionally, acquiring the lock to remove the transaction
from its list *only* when the count reaches 0. This does not
protect the quiesce path, which depends on the lock to ensure its
extra reference prevents release of the transaction.
Fix this by only dropping the last reference to a transaction
in gsi_trans_free() while holding the spinlock.
Fixes: 9dd441e4ed575 ("soc: qcom: ipa: GSI transactions") Reported-by: Stephen Boyd <swboyd@chromium.org> Signed-off-by: Alex Elder <elder@linaro.org> Link: https://lore.kernel.org/r/20201114182017.28270-1-elder@linaro.org Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
DSA network devices rely on having their DSA management interface up and
running otherwise their ndo_open() will return -ENETDOWN. Without doing
this it would not be possible to use DSA devices as netconsole when
configured on the command line. These devices also do not utilize the
upper/lower linking so the check about the netpoll device having upper
is not going to be a problem.
The solution adopted here is identical to the one done for
net/ipv4/ipconfig.c with 728c02089a0e ("net: ipv4: handle DSA enabled
master network devices"), with the network namespace scope being
restricted to that of the process configuring netpoll.
When removing the driver we would hit BUG_ON(!list_empty(&dev->ptype_specific))
in net/core/dev.c due to still having the NC-SI packet handler
registered.
Fixes: bd466c3fb5a4 ("net/faraday: Support NCSI mode") Signed-off-by: Joel Stanley <joel@jms.id.au> Link: https://lore.kernel.org/r/20201117024448.1170761-1-joel@jms.id.au Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Depending on the SoC/platform the CPSW can completely lose context after a
suspend/resume cycle, including CPSW wrapper (WR) which will cause reset of
WR_C0_MISC_EN register, so CPTS IRQ will became disabled.
Fix it by moving CPTS IRQ enabling in cpsw_ndo_open() where CPTS is
actually started.
The ndo_start_xmit() method must return NETDEV_TX_OK if the DMA mapping
fails, after freeing the socket buffer.
Fix the mtk_star_netdev_start_xmit() function accordingly.
Fixes: 8c7bd5a454ff ("net: ethernet: mtk-star-emac: new driver") Signed-off-by: Vincent Stehlé <vincent.stehle@laposte.net> Acked-by: Bartosz Golaszewski <bgolaszewski@baylibre.com> Link: https://lore.kernel.org/r/20201112084833.21842-1-vincent.stehle@laposte.net Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
As soon as you add the second port to a VLAN, all other port
membership configuration is overwritten with zeroes. The HW interprets
this as all ports being "unmodified members" of the VLAN.
In the simple case when all ports belong to the same VLAN, switching
will still work. But using multiple VLANs or trying to set multiple
ports as tagged members will not work.
On the 6352, doing a VTU GetNext op, followed by an STU GetNext op
will leave you with both the member- and state- data in the VTU/STU
data registers. But on the 6097 (which uses the same implementation),
the STU GetNext will override the information gathered from the VTU
GetNext.
Separate the two stages, parsing the result of the VTU GetNext before
doing the STU GetNext.
We opt to update the existing implementation for all applicable chips,
as opposed to creating a separate callback for 6097, because although
the previous implementation did work for (at least) 6352, the
datasheet does not mention the masking behavior.
Fixes: ef6fcea37f01 ("net: dsa: mv88e6xxx: get STU entry on VTU GetNext") Signed-off-by: Tobias Waldekranz <tobias@waldekranz.com> Link: https://lore.kernel.org/r/20201112114335.27371-1-tobias@waldekranz.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Fix to return a negative error code from the error handling
case instead of 0, as done elsewhere in this function.
Fixes: 39a6f4bce6b4 ("b44: replace the ssb_dma API with the generic DMA API") Reported-by: Hulk Robot <hulkci@huawei.com> Signed-off-by: Zhang Changzhong <zhangchangzhong@huawei.com> Reviewed-by: Michael Chan <michael.chan@broadcom.com> Link: https://lore.kernel.org/r/1605582131-36735-1-git-send-email-zhangchangzhong@huawei.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The driver sends Ethernet Management Datagram (EMAD) packets to the
device for configuration purposes and waits for up to 200ms for a reply.
A request is retried up to 5 times.
When the system is under heavy load, replies are not always processed in
time and EMAD transactions fail.
Make the process more robust to such delays by using exponential
backoff. First wait for up to 200ms, then retransmit and wait for up to
400ms and so on.
On arm imx6, when opening the chip's netdev, the whole Linux
kernel intermittently hangs/freezes.
This is caused by a bug in the driver code which tests if pcie
interrupts are working correctly, using the software interrupt:
1. open: enable the software interrupt
2. open: tell the chip to assert the software interrupt
3. open: wait for flag
4. ISR: acknowledge s/w interrupt, set flag
5. open: notice flag, disable the s/w interrupt, continue
Unfortunately the ISR only acknowledges the s/w interrupt, but
does not disable it. This will re-trigger the ISR in a tight
loop.
On some (lucky) platforms, open proceeds to disable the s/w
interrupt even while the ISR is 'spinning'. On arm imx6,
the spinning ISR does not allow open to proceed, resulting
in a hung Linux kernel.
Fix minimally by disabling the s/w interrupt in the ISR, which
will prevent it from spinning. This won't break anything because
the s/w interrupt is used as a one-shot interrupt.
Note that this is a minimal fix, overlooking many possible
cleanups, e.g.:
- lan743x_intr_software_isr() is completely redundant and reads
INT_STS twice for no apparent reason
- disabling the s/w interrupt in lan743x_intr_test_isr() is now
redundant, but harmless
- waiting on software_isr_flag can be converted from a sleeping
poll loop to wait_event_timeout()
Fixes: 23f0703c125b ("lan743x: Add main source files for new lan743x driver") Tested-by: Sven Van Asbroeck <thesven73@gmail.com> # arm imx6 lan7430 Signed-off-by: Sven Van Asbroeck <thesven73@gmail.com> Link: https://lore.kernel.org/r/20201112204741.12375-1-TheSven73@gmail.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When running this chip on arm imx6, we intermittently observe
the following kernel warning in the log, especially when the
system is under high load:
[ 50.119484] ------------[ cut here ]------------
[ 50.124377] WARNING: CPU: 0 PID: 303 at kernel/softirq.c:169 __local_bh_enable_ip+0x100/0x184
[ 50.132925] IRQs not enabled as expected
[ 50.159250] CPU: 0 PID: 303 Comm: rngd Not tainted 5.7.8 #1
[ 50.164837] Hardware name: Freescale i.MX6 Quad/DualLite (Device Tree)
[ 50.171395] [<c0111a38>] (unwind_backtrace) from [<c010be28>] (show_stack+0x10/0x14)
[ 50.179162] [<c010be28>] (show_stack) from [<c05b9dec>] (dump_stack+0xac/0xd8)
[ 50.186408] [<c05b9dec>] (dump_stack) from [<c0122e40>] (__warn+0xd0/0x10c)
[ 50.193391] [<c0122e40>] (__warn) from [<c0123238>] (warn_slowpath_fmt+0x98/0xc4)
[ 50.200892] [<c0123238>] (warn_slowpath_fmt) from [<c012b010>] (__local_bh_enable_ip+0x100/0x184)
[ 50.209860] [<c012b010>] (__local_bh_enable_ip) from [<bf09ecbc>] (destroy_conntrack+0x48/0xd8 [nf_conntrack])
[ 50.220038] [<bf09ecbc>] (destroy_conntrack [nf_conntrack]) from [<c0ac9b58>] (nf_conntrack_destroy+0x94/0x168)
[ 50.230160] [<c0ac9b58>] (nf_conntrack_destroy) from [<c0a4aaa0>] (skb_release_head_state+0xa0/0xd0)
[ 50.239314] [<c0a4aaa0>] (skb_release_head_state) from [<c0a4aadc>] (skb_release_all+0xc/0x24)
[ 50.247946] [<c0a4aadc>] (skb_release_all) from [<c0a4b4cc>] (consume_skb+0x74/0x17c)
[ 50.255796] [<c0a4b4cc>] (consume_skb) from [<c081a2dc>] (lan743x_tx_release_desc+0x120/0x124)
[ 50.264428] [<c081a2dc>] (lan743x_tx_release_desc) from [<c081a98c>] (lan743x_tx_napi_poll+0x5c/0x18c)
[ 50.273755] [<c081a98c>] (lan743x_tx_napi_poll) from [<c0a6b050>] (net_rx_action+0x118/0x4a4)
[ 50.282306] [<c0a6b050>] (net_rx_action) from [<c0101364>] (__do_softirq+0x13c/0x53c)
[ 50.290157] [<c0101364>] (__do_softirq) from [<c012b29c>] (irq_exit+0x150/0x17c)
[ 50.297575] [<c012b29c>] (irq_exit) from [<c0196a08>] (__handle_domain_irq+0x60/0xb0)
[ 50.305423] [<c0196a08>] (__handle_domain_irq) from [<c05d44fc>] (gic_handle_irq+0x4c/0x90)
[ 50.313790] [<c05d44fc>] (gic_handle_irq) from [<c0100ed4>] (__irq_usr+0x54/0x80)
[ 50.321287] Exception stack(0xecd99fb0 to 0xecd99ff8)
[ 50.326355] 9fa0: 1cf1aa74000000010000000100000000
[ 50.334547] 9fc0: 00000001000000000000000000000000000000000000000000004097b6d17d14
[ 50.342738] 9fe0: 00000001b6d17c6000000000b6e71f94800b0010ffffffff
[ 50.349364] irq event stamp: 2525027
[ 50.352955] hardirqs last enabled at (2525026): [<c0a6afec>] net_rx_action+0xb4/0x4a4
[ 50.360892] hardirqs last disabled at (2525027): [<c0d6d2fc>] _raw_spin_lock_irqsave+0x1c/0x50
[ 50.369517] softirqs last enabled at (2524660): [<c01015b4>] __do_softirq+0x38c/0x53c
[ 50.377446] softirqs last disabled at (2524693): [<c012b29c>] irq_exit+0x150/0x17c
[ 50.385027] ---[ end trace c0b571db4bc8087d ]---
The driver is calling dev_kfree_skb() from code inside a spinlock,
where h/w interrupts are disabled. This is forbidden, as documented
in include/linux/netdevice.h. The correct function to use
dev_kfree_skb_irq(), or dev_kfree_skb_any().
Fix by using the correct dev_kfree_skb_xxx() functions:
in lan743x_tx_release_desc():
called by lan743x_tx_release_completed_descriptors()
called by in lan743x_tx_napi_poll()
which holds a spinlock
called by lan743x_tx_release_all_descriptors()
called by lan743x_tx_close()
which can-sleep
conclusion: use dev_kfree_skb_any()
in lan743x_tx_xmit_frame():
which holds a spinlock
conclusion: use dev_kfree_skb_irq()
in lan743x_tx_close():
which can-sleep
conclusion: use dev_kfree_skb()
in lan743x_rx_release_ring_element():
called by lan743x_rx_close()
which can-sleep
called by lan743x_rx_open()
which can-sleep
conclusion: use dev_kfree_skb()
Fixes: 23f0703c125b ("lan743x: Add main source files for new lan743x driver") Signed-off-by: Sven Van Asbroeck <thesven73@gmail.com> Link: https://lore.kernel.org/r/20201112185949.11315-1-TheSven73@gmail.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
nlmsg_cancel() needs to be called in the error path of
inet_req_diag_fill to cancel the message.
Fixes: d545caca827b ("net: inet: diag: expose the socket mark to privileged processes.") Reported-by: Hulk Robot <hulkci@huawei.com> Signed-off-by: Wang Hai <wanghai38@huawei.com> Link: https://lore.kernel.org/r/20201116082018.16496-1-wanghai38@huawei.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit 58956317c8de ("neighbor: Improve garbage collection")
guarantees neighbour table entries a five-second lifetime. Processes
which make heavy use of multicast can fill the neighour table with
multicast addresses in five seconds. At that point, neighbour entries
can't be GC-ed because they aren't five seconds old yet, the kernel
log starts to fill up with "neighbor table overflow!" messages, and
sends start to fail.
This patch allows multicast addresses to be thrown out before they've
lived out their five seconds. This makes room for non-multicast
addresses and makes messages to all addresses more reliable in these
circumstances.
Fixes: 58956317c8de ("neighbor: Improve garbage collection") Signed-off-by: Jeff Dike <jdike@akamai.com> Reviewed-by: David Ahern <dsahern@kernel.org> Link: https://lore.kernel.org/r/20201113015815.31397-1-jdike@akamai.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
projects / thirdparty / kernel / stable.git / log
