Merge tag 'copy-struct-from-user-v5.4-rc4' of gitolite.kernel.org:pub/scm/linux/kernel/git/brauner/linux

Pull usercopy test fixlets from Christian Brauner:
 "This contains two improvements for the copy_struct_from_user() tests:

   - a coding style change to get rid of the ugly "if ((ret |= test()))"
     pointed out when pulling the original patchset.

   - avoid a soft lockups when running the usercopy tests on machines
     with large page sizes by scanning only a 1024 byte region"

* tag 'copy-struct-from-user-v5.4-rc4' of gitolite.kernel.org:pub/scm/linux/kernel/git/brauner/linux:
  usercopy: Avoid soft lockups in test_check_nonzero_user()
  lib: test_user_copy: style cleanup

selftests/bpf: More compatible nc options in test_tc_edt

Out of the three nc implementations widely in use, at least two (BSD netcat
and nmap-ncat) do not support -l combined with -s. Modify the nc invocation
to be accepted by all of them.

Fixes: 7df5e3db8f63 ("selftests: bpf: tc-bpf flow shaping with EDT")
Signed-off-by: Jiri Benc <[email protected]>
Signed-off-by: Daniel Borkmann <[email protected]>
Acked-by: Peter Oskolkov <[email protected]>
Link: https://lore.kernel.org/bpf/f5bf07dccd8b552a76c84d49e80b86c5aa071122.1571400024.git.jbenc@redhat.com

net/mlx5: fix memory leak in mlx5_fw_fatal_reporter_dump

In mlx5_fw_fatal_reporter_dump if mlx5_crdump_collect fails the
allocated memory for cr_data must be released otherwise there will be
memory leak. To fix this, this commit changes the return instruction
into goto error handling.

Fixes: 9b1f29823605 ("net/mlx5: Add support for FW fatal reporter dump")
Signed-off-by: Navid Emamdoost <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5: prevent memory leak in mlx5_fpga_conn_create_cq

In mlx5_fpga_conn_create_cq if mlx5_vector2eqn fails the allocated
memory should be released.

Fixes: 537a50574175 ("net/mlx5: FPGA, Add high-speed connection routines")
Signed-off-by: Navid Emamdoost <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: TX, Fix consumer index of error cqe dump

The completion queue consumer index increments upon a call to
mlx5_cqwq_pop().
When dumping an error CQE, the index is already incremented.
Decrease one for the print command.

Fixes: 16cc14d81733 ("net/mlx5e: Dump xmit error completions")
Signed-off-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: kTLS, Enhance TX resync flow

Once the kTLS TX resync function is called, it used to return
a binary value, for success or failure.

However, in case the TLS SKB is a retransmission of the connection
handshake, it initiates the resync flow (as the tcp seq check holds),
while regular packet handle is expected.

In this patch, we identify this case and skip the resync operation
accordingly.

Counters:
- Add a counter (tls_skip_no_sync_data) to monitor this.
- Bump the dump counters up as they are used more frequently.
- Add a missing counter descriptor declaration for tls_resync_bytes
  in sq_stats_desc.

Fixes: d2ead1f360e8 ("net/mlx5e: Add kTLS TX HW offload support")
Signed-off-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: kTLS, Save a copy of the crypto info

Do not assume the crypto info is accessible during the
connection lifetime. Save a copy of it in the private
TX context.

Fixes: d2ead1f360e8 ("net/mlx5e: Add kTLS TX HW offload support")
Signed-off-by: Tariq Toukan <[email protected]>
Reviewed-by: Eran Ben Elisha <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: kTLS, Remove unneeded cipher type checks

Cipher type is checked upon connection addition.
No need to recheck it per every TX resync invocation.

Fixes: d2ead1f360e8 ("net/mlx5e: Add kTLS TX HW offload support")
Signed-off-by: Tariq Toukan <[email protected]>
Reviewed-by: Eran Ben Elisha <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: kTLS, Limit DUMP wqe size

HW expects the data size in DUMP WQEs to be up to MTU.
Make sure they are in range.

We elevate the frag page refcount by 'n-1', in addition to the
one obtained in tx_sync_info_get(), having an overall of 'n'
references. We bulk increments by using a single page_ref_add()
command, to optimize perfermance.
The refcounts are released one by one, by the corresponding completions.

Fixes: d2ead1f360e8 ("net/mlx5e: Add kTLS TX HW offload support")
Signed-off-by: Tariq Toukan <[email protected]>
Reviewed-by: Eran Ben Elisha <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: kTLS, Fix missing SQ edge fill

Before posting the context params WQEs, make sure there is enough
contiguous room for them, and fill frag edge if needed.

When posting only a nop, no need for room check, as it needs a single
WQEBB, meaning no contiguity issue.

Fixes: d2ead1f360e8 ("net/mlx5e: Add kTLS TX HW offload support")
Signed-off-by: Tariq Toukan <[email protected]>
Reviewed-by: Eran Ben Elisha <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: kTLS, Fix page refcnt leak in TX resync error flow

All references for frag pages that are obtained in tx_sync_info_get()
should be released.
Release usually occurs in the corresponding CQE of the WQE.
In error flows, not all fragments have a WQE posted for them, hence
no matching CQE will be generated.
For these pages, release the reference in the error flow.

Fixes: d2ead1f360e8 ("net/mlx5e: Add kTLS TX HW offload support")
Signed-off-by: Tariq Toukan <[email protected]>
Reviewed-by: Eran Ben Elisha <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: kTLS, Save by-value copy of the record frags

Access the record fragments only under the TLS ctx lock.
In the resync flow, save a copy of them to be used when
preparing and posting the required DUMP WQEs.

Fixes: d2ead1f360e8 ("net/mlx5e: Add kTLS TX HW offload support")
Signed-off-by: Tariq Toukan <[email protected]>
Reviewed-by: Eran Ben Elisha <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: kTLS, Save only the frag page to release at completion

In TX resync flow where DUMP WQEs are posted, keep a pointer to
the fragment page to unref it upon completion, instead of saving
the whole fragment.

In addition, move it the end of the arguments list in tx_fill_wi().

Fixes: d2ead1f360e8 ("net/mlx5e: Add kTLS TX HW offload support")
Signed-off-by: Tariq Toukan <[email protected]>
Reviewed-by: Eran Ben Elisha <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: kTLS, Size of a Dump WQE is fixed

No Eth segment, so no dynamic inline headers.
The size of a Dump WQE is fixed, use constants and remove
unnecessary checks.

Fixes: d2ead1f360e8 ("net/mlx5e: Add kTLS TX HW offload support")
Signed-off-by: Tariq Toukan <[email protected]>
Reviewed-by: Eran Ben Elisha <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: kTLS, Release reference on DUMPed fragments in shutdown flow

A call to kTLS completion handler was missing in the TXQSQ release
flow. Add it.

Fixes: d2ead1f360e8 ("net/mlx5e: Add kTLS TX HW offload support")
Signed-off-by: Tariq Toukan <[email protected]>
Reviewed-by: Eran Ben Elisha <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: Tx, Zero-memset WQE info struct upon update

Not all fields of WQE info are being written in the function,
having some leftovers from previous rounds.
Zero-memset it upon update.

Particularly, not nullifying the wi->resync_dump_frag field
will cause double free of the kTLS DUMPed frags.

Fixes: d2ead1f360e8 ("net/mlx5e: Add kTLS TX HW offload support")
Signed-off-by: Tariq Toukan <[email protected]>
Reviewed-by: Eran Ben Elisha <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: Tx, Fix assumption of single WQEBB of NOP in cleanup flow

Cited patch removed the assumption only in datapath.
Here we remove it also form control/cleanup flow.

Fixes: 9ab0233728ca ("net/mlx5e: Tx, Don't implicitly assume SKB-less wqe has one WQEBB")
Signed-off-by: Tariq Toukan <[email protected]>
Reviewed-by: Eran Ben Elisha <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net: usb: lan78xx: Connect PHY before registering MAC

As soon as the netdev is registers, the kernel can start using the
interface. If the driver connects the MAC to the PHY after the netdev
is registered, there is a race condition where the interface can be
opened without having the PHY connected.

Change the order to close this race condition.

Fixes: 92571a1aae40 ("lan78xx: Connect phy early")
Reported-by: Daniel Wagner <[email protected]>
Signed-off-by: Andrew Lunn <[email protected]>
Tested-by: Daniel Wagner <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge branch 'vsock-virtio-make-the-credit-mechanism-more-robust'

Stefano Garzarella says:

====================
vsock/virtio: make the credit mechanism more robust

This series makes the credit mechanism implemented in the
virtio-vsock devices more robust.
Patch 1 sends an update to the remote peer when the buf_alloc
change.
Patch 2 prevents a malicious peer (especially the guest) can
consume all the memory of the other peer, discarding packets
when the credit available is not respected.
====================

Signed-off-by: David S. Miller <[email protected]>

vsock/virtio: discard packets if credit is not respected

If the remote peer doesn't respect the credit information
(buf_alloc, fwd_cnt), sending more data than it can send,
we should drop the packets to prevent a malicious peer
from using all of our memory.

This is patch follows the VIRTIO spec: "VIRTIO_VSOCK_OP_RW data
packets MUST only be transmitted when the peer has sufficient
free buffer space for the payload"

Signed-off-by: Stefano Garzarella <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

vsock/virtio: send a credit update when buffer size is changed

When the user application set a new buffer size value, we should
update the remote peer about this change, since it uses this
information to calculate the credit available.

Signed-off-by: Stefano Garzarella <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

mlxsw: spectrum_trap: Push Ethernet header before reporting trap

devlink maintains packets and bytes statistics for each trap. Since
eth_type_trans() was called to set the skb's protocol, the data pointer
no longer points to the start of the packet and the bytes accounting is
off by 14 bytes.

Fix this by pushing the skb's data pointer to the start of the packet.

Fixes: b5ce611fd96e ("mlxsw: spectrum: Add devlink-trap support")
Reported-by: Alex Kushnarov <[email protected]>
Tested-by: Alex Kushnarov <[email protected]>
Acked-by: Jiri Pirko <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

net: ensure correct skb->tstamp in various fragmenters

Thomas found that some forwarded packets would be stuck
in FQ packet scheduler because their skb->tstamp contained
timestamps far in the future.

We thought we addressed this point in commit 8203e2d844d3
("net: clear skb->tstamp in forwarding paths") but there
is still an issue when/if a packet needs to be fragmented.

In order to meet EDT requirements, we have to make sure all
fragments get the original skb->tstamp.

Note that this original skb->tstamp should be zero in
forwarding path, but might have a non zero value in
output path if user decided so.

Fixes: fb420d5d91c1 ("tcp/fq: move back to CLOCK_MONOTONIC")
Signed-off-by: Eric Dumazet <[email protected]>
Reported-by: Thomas Bartschies <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge tag 'mmc-v5.4-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc

Pull MMC fixes from Ulf Hansson:
 "MMC host:
   - sdhci-iproc: Prevent some spurious interrupts
   - renesas_sdhi/sh_mmcif: Avoid false warnings about IRQs not found

  MEMSTICK host:
   - jmb38x_ms: Fix an error handling path at ->probe()"

* tag 'mmc-v5.4-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc:
  memstick: jmb38x_ms: Fix an error handling path in 'jmb38x_ms_probe()'
  mmc: sdhci-iproc: fix spurious interrupts on Multiblock reads with bcm2711
  mmc: sh_mmcif: Use platform_get_irq_optional() for optional interrupt
  mmc: renesas_sdhi: Do not use platform_get_irq() to count interrupts

Merge branch 'net-bcmgenet-restore-internal-EPHY-support'

Doug Berger says:

====================
net: bcmgenet: restore internal EPHY support

I managed to get my hands on an old BCM97435SVMB board to do some
testing with the latest kernel and uncovered a number of things
that managed to get broken over the years (some by me ;).

This commit set attempts to correct the errors I observed in my
testing.

The first commit applies to all internal PHYs to restore proper
reporting of link status when a link comes up.

The second commit restores the soft reset to the initialization of
the older internal EPHYs used by 40nm Set-Top Box devices.

The third corrects a bug I introduced when removing excessive soft
resets by altering the initialization sequence in a way that keeps
the GENETv3 MAC interface happy.

Finally, I observed a number of issues when manually configuring
the network interface of the older EPHYs that appear to be resolved
by the fourth commit.
====================

Signed-off-by: David S. Miller <[email protected]>

net: bcmgenet: reset 40nm EPHY on energy detect

The EPHY integrated into the 40nm Set-Top Box devices can falsely
detect energy when connected to a disabled peer interface. When the
peer interface is enabled the EPHY will detect and report the link
as active, but on occasion may get into a state where it is not
able to exchange data with the connected GENET MAC. This issue has
not been observed when the link parameters are auto-negotiated;
however, it has been observed with a manually configured link.

It has been empirically determined that issuing a soft reset to the
EPHY when energy is detected prevents it from getting into this bad
state.

Fixes: 1c1008c793fa ("net: bcmgenet: add main driver file")
Signed-off-by: Doug Berger <[email protected]>
Acked-by: Florian Fainelli <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

net: bcmgenet: soft reset 40nm EPHYs before MAC init

It turns out that the "Workaround for putting the PHY in IDDQ mode"
used by the internal EPHYs on 40nm Set-Top Box chips when powering
down puts the interface to the GENET MAC in a state that can cause
subsequent MAC resets to be incomplete.

Rather than restore the forced soft reset when powering up internal
PHYs, this commit moves the invocation of phy_init_hw earlier in
the MAC initialization sequence to just before the MAC reset in the
open and resume functions. This allows the interface to be stable
and allows the MAC resets to be successful.

The bcmgenet_mii_probe() function is split in two to accommodate
this. The new function bcmgenet_mii_connect() handles the first
half of the functionality before the MAC initialization, and the
bcmgenet_mii_config() function is extended to provide the remaining
PHY configuration following the MAC initialization.

Fixes: 484bfa1507bf ("Revert "net: bcmgenet: Software reset EPHY after power on"")
Signed-off-by: Doug Berger <[email protected]>
Acked-by: Florian Fainelli <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

net: phy: bcm7xxx: define soft_reset for 40nm EPHY

The internal 40nm EPHYs use a "Workaround for putting the PHY in
IDDQ mode." These PHYs require a soft reset to restore functionality
after they are powered back up.

This commit defines the soft_reset function to use genphy_soft_reset
during phy_init_hw to accommodate this.

Fixes: 6e2d85ec0559 ("net: phy: Stop with excessive soft reset")
Signed-off-by: Doug Berger <[email protected]>
Acked-by: Florian Fainelli <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

net: bcmgenet: don't set phydev->link from MAC

When commit 28b2e0d2cd13 ("net: phy: remove parameter new_link from
phy_mac_interrupt()") removed the new_link parameter it set the
phydev->link state from the MAC before invoking phy_mac_interrupt().

However, once commit 88d6272acaaa ("net: phy: avoid unneeded MDIO
reads in genphy_read_status") was added this initialization prevents
the proper determination of the connection parameters by the function
genphy_read_status().

This commit removes that initialization to restore the proper
functionality.

Fixes: 88d6272acaaa ("net: phy: avoid unneeded MDIO reads in genphy_read_status")
Signed-off-by: Doug Berger <[email protected]>
Acked-by: Florian Fainelli <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge tag 'sound-5.4-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound

Pull sound fixes from Takashi Iwai:
 "Just a few small fixes for the usual suspect, HD- and USB-audio:
  enablement of runtime PM for Nvidia due to the recent PCI changes, a
  fix for potential hangs with recent HD-audio platforms, and the rest
  device-specific quirks"

* tag 'sound-5.4-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound:
  ALSA: hda - Force runtime PM on Nvidia HDMI codecs
  ALSA: hda/realtek - Enable headset mic on Asus MJ401TA
  ALSA: usb-audio: Disable quirks for BOSS Katana amplifiers
  ALSA: hdac: clear link output stream mapping
  ALSA: hda/realtek: Reduce the Headphone static noise on XPS 9350/9360

Merge tag 'acpi-5.4-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm

Pull ACPI fixes from Rafael Wysocki:
 "Fix possible use-after-free in the ACPI CPPC support code (John Garry)
  and prevent the ACPI HMAT parsing code from using possibly incorrect
  data coming from the platform firmware (Daniel Black)"

* tag 'acpi-5.4-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  ACPI: CPPC: Set pcc_data[pcc_ss_id] to NULL in acpi_cppc_processor_exit()
  ACPI: HMAT: ACPI_HMAT_MEMORY_PD_VALID is deprecated since ACPI-6.3

Merge tag 'pm-5.4-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm

Pull power management fixes from Rafael Wysocki:
 "These include a fix for a recent regression in the ACPI CPU
performance scaling code, a PCI device power management fix,
a system shutdown fix related to cpufreq, a removal of an ACPI
suspend-to-idle blacklist entry and a build warning fix.

Specifics:

   - Fix possible NULL pointer dereference in the ACPI processor scaling
     initialization code introduced by a recent cpufreq update (Rafael
     Wysocki).

   - Fix possible deadlock due to suspending cpufreq too late during
     system shutdown (Rafael Wysocki).

   - Make the PCI device system resume code path be more consistent with
     its PM-runtime counterpart to fix an issue with missing delay on
     transitions from D3cold to D0 during system resume from
     suspend-to-idle on some systems (Rafael Wysocki).

   - Drop Dell XPS13 9360 from the LPS0 Idle _DSM blacklist to make it
     use suspend-to-idle by default (Mario Limonciello).

   - Fix build warning in the core system suspend support code (Ben
     Dooks)"

* tag 'pm-5.4-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  ACPI: processor: Avoid NULL pointer dereferences at init time
  PCI: PM: Fix pci_power_up()
  PM: sleep: include <linux/pm_runtime.h> for pm_wq
  cpufreq: Avoid cpufreq_suspend() deadlock on system shutdown
  ACPI: PM: Drop Dell XPS13 9360 from LPS0 Idle _DSM blacklist

Merge tag 'mkp-scsi-postmerge' of git://git.kernel.org/pub/scm/linux/kernel/git/mkp/scsi

Pull scsi fixes from Martin Petersen:
 "These two commits were in a separate postmerge branch due to a
  dependency on changes merged for 5.4 in the block tree.

  They fix two issues in the intersection of the request cleanup changes
  from block (b7e9e1fb7a92) and the request batching changes
  (8930a6c20791) that were made to SCSI during the 5.4 cycle"

* tag 'mkp-scsi-postmerge' of git://git.kernel.org/pub/scm/linux/kernel/git/mkp/scsi:
  scsi: core: fix dh and multipathing for SCSI hosts without request batching
  scsi: core: fix missing .cleanup_rq for SCSI hosts without request batching

iommu/amd: Check PM_LEVEL_SIZE() condition in locked section

The increase_address_space() function has to check the PM_LEVEL_SIZE()
condition again under the domain->lock to avoid a false trigger of the
WARN_ON_ONCE() and to avoid that the address space is increase more
often than necessary.

Reported-by: Qian Cai <[email protected]>
Fixes: 754265bcab78 ("iommu/amd: Fix race in increase_address_space()")
Reviewed-by: Jerry Snitselaar <[email protected]>
Signed-off-by: Joerg Roedel <[email protected]>

Merge branch 'nvme-5.4' of git://git.infradead.org/nvme into for-linus

Pull NVMe updates from Keith:

"This is a collection of bug fixes committed since the previous pull
 request that address deadlocks, double resets, memory leaks, and other
 regression."

* 'nvme-5.4' of git://git.infradead.org/nvme:
  nvme-pci: Set the prp2 correctly when using more than 4k page
  nvme-tcp: fix possible leakage during error flow
  nvmet-loop: fix possible leakage during error flow
  nvme-tcp: Initialize sk->sk_ll_usec only with NET_RX_BUSY_POLL
  nvme: Wait for reset state when required
  nvme: Prevent resets during paused controller state
  nvme: Restart request timers in resetting state
  nvme: Remove ADMIN_ONLY state
  nvme-pci: Free tagset if no IO queues
  nvme: retain split access workaround for capability reads
  nvme: fix possible deadlock when nvme_update_formats fails

nvme-pci: Set the prp2 correctly when using more than 4k page

In the current code, the nvme is using a fixed 4k PRP entry size,
but if the kernel use a page size which is more than 4k, we should
consider the situation that the bv_offset may be larger than the
dev->ctrl.page_size. Otherwise we may miss setting the prp2 and then
cause the command can't be executed correctly.

Fixes: dff824b2aadb ("nvme-pci: optimize mapping of small single segment requests")
Cc: [email protected]
Reviewed-by: Christoph Hellwig <[email protected]>
Signed-off-by: Kevin Hao <[email protected]>
Signed-off-by: Keith Busch <[email protected]>

Merge branch 'acpi-tables'

* acpi-tables:
  ACPI: HMAT: ACPI_HMAT_MEMORY_PD_VALID is deprecated since ACPI-6.3

ACPI: CPPC: Set pcc_data[pcc_ss_id] to NULL in acpi_cppc_processor_exit()

When enabling KASAN and DEBUG_TEST_DRIVER_REMOVE, I find this KASAN
warning:

[   20.872057] BUG: KASAN: use-after-free in pcc_data_alloc+0x40/0xb8
[   20.878226] Read of size 4 at addr ffff00236cdeb684 by task swapper/0/1
[   20.884826]
[   20.886309] CPU: 19 PID: 1 Comm: swapper/0 Not tainted 5.4.0-rc1-00009-ge7f7df3db5bf-dirty #289
[   20.894994] Hardware name: Huawei D06 /D06, BIOS Hisilicon D06 UEFI RC0 - V1.16.01 03/15/2019
[   20.903505] Call trace:
[   20.905942]  dump_backtrace+0x0/0x200
[   20.909593]  show_stack+0x14/0x20
[   20.912899]  dump_stack+0xd4/0x130
[   20.916291]  print_address_description.isra.9+0x6c/0x3b8
[   20.921592]  __kasan_report+0x12c/0x23c
[   20.925417]  kasan_report+0xc/0x18
[   20.928808]  __asan_load4+0x94/0xb8
[   20.932286]  pcc_data_alloc+0x40/0xb8
[   20.935938]  acpi_cppc_processor_probe+0x4e8/0xb08
[   20.940717]  __acpi_processor_start+0x48/0xb0
[   20.945062]  acpi_processor_start+0x40/0x60
[   20.949235]  really_probe+0x118/0x548
[   20.952887]  driver_probe_device+0x7c/0x148
[   20.957059]  device_driver_attach+0x94/0xa0
[   20.961231]  __driver_attach+0xa4/0x110
[   20.965055]  bus_for_each_dev+0xe8/0x158
[   20.968966]  driver_attach+0x30/0x40
[   20.972531]  bus_add_driver+0x234/0x2f0
[   20.976356]  driver_register+0xbc/0x1d0
[   20.980182]  acpi_processor_driver_init+0x40/0xe4
[   20.984875]  do_one_initcall+0xb4/0x254
[   20.988700]  kernel_init_freeable+0x24c/0x2f8
[   20.993047]  kernel_init+0x10/0x118
[   20.996524]  ret_from_fork+0x10/0x18
[   21.000087]
[   21.001567] Allocated by task 1:
[   21.004785]  save_stack+0x28/0xc8
[   21.008089]  __kasan_kmalloc.isra.9+0xbc/0xd8
[   21.012435]  kasan_kmalloc+0xc/0x18
[   21.015913]  pcc_data_alloc+0x94/0xb8
[   21.019564]  acpi_cppc_processor_probe+0x4e8/0xb08
[   21.024343]  __acpi_processor_start+0x48/0xb0
[   21.028689]  acpi_processor_start+0x40/0x60
[   21.032860]  really_probe+0x118/0x548
[   21.036512]  driver_probe_device+0x7c/0x148
[   21.040684]  device_driver_attach+0x94/0xa0
[   21.044855]  __driver_attach+0xa4/0x110
[   21.048680]  bus_for_each_dev+0xe8/0x158
[   21.052591]  driver_attach+0x30/0x40
[   21.056155]  bus_add_driver+0x234/0x2f0
[   21.059980]  driver_register+0xbc/0x1d0
[   21.063805]  acpi_processor_driver_init+0x40/0xe4
[   21.068497]  do_one_initcall+0xb4/0x254
[   21.072322]  kernel_init_freeable+0x24c/0x2f8
[   21.076667]  kernel_init+0x10/0x118
[   21.080144]  ret_from_fork+0x10/0x18
[   21.083707]
[   21.085186] Freed by task 1:
[   21.088056]  save_stack+0x28/0xc8
[   21.091360]  __kasan_slab_free+0x118/0x180
[   21.095445]  kasan_slab_free+0x10/0x18
[   21.099183]  kfree+0x80/0x268
[   21.102139]  acpi_cppc_processor_exit+0x1a8/0x1b8
[   21.106832]  acpi_processor_stop+0x70/0x80
[   21.110917]  really_probe+0x174/0x548
[   21.114568]  driver_probe_device+0x7c/0x148
[   21.118740]  device_driver_attach+0x94/0xa0
[   21.122912]  __driver_attach+0xa4/0x110
[   21.126736]  bus_for_each_dev+0xe8/0x158
[   21.130648]  driver_attach+0x30/0x40
[   21.134212]  bus_add_driver+0x234/0x2f0
[   21.0x10/0x18
[   21.161764]
[   21.163244] The buggy address belongs to the object at ffff00236cdeb600
[   21.163244]  which belongs to the cache kmalloc-256 of size 256
[   21.175750] The buggy address is located 132 bytes inside of
[   21.175750]  256-byte region [ffff00236cdeb600, ffff00236cdeb700)
[   21.187473] The buggy address belongs to the page:
[   21.192254] page:fffffe008d937a00 refcount:1 mapcount:0 mapping:ffff002370c0fa00 index:0x0 compound_mapcount: 0
[   21.202331] flags: 0x1ffff00000010200(slab|head)
[   21.206940] raw: 1ffff00000010200 dead000000000100 dead000000000122 ffff002370c0fa00
[   21.214671] raw: 0000000000000000 00000000802a002a 00000001ffffffff 0000000000000000
[   21.222400] page dumped because: kasan: bad access detected
[   21.227959]
[   21.229438] Memory state around the buggy address:
[   21.234218]  ffff00236cdeb580: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
[   21.241427]  ffff00236cdeb600: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[   21.248637] >ffff00236cdeb680: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[   21.255845]                    ^
[   21.259062]  ffff00236cdeb700: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
[   21.266272]  ffff00236cdeb780: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[   21.273480] ==================================================================

It seems that global pcc_data[pcc_ss_id] can be freed in
acpi_cppc_processor_exit(), but we may later reference this value, so
NULLify it when freed.

Also remove the useless setting of data "pcc_channel_acquired", which
we're about to free.

Fixes: 85b1407bf6d2 ("ACPI / CPPC: Make CPPC ACPI driver aware of PCC subspace IDs")
Signed-off-by: John Garry <[email protected]>
Cc: 4.15+ <[email protected]> # 4.15+
Signed-off-by: Rafael J. Wysocki <[email protected]>

Merge branches 'pm-cpufreq' and 'pm-sleep'

* pm-cpufreq:
  ACPI: processor: Avoid NULL pointer dereferences at init time
  cpufreq: Avoid cpufreq_suspend() deadlock on system shutdown

* pm-sleep:
  PM: sleep: include <linux/pm_runtime.h> for pm_wq
  ACPI: PM: Drop Dell XPS13 9360 from LPS0 Idle _DSM blacklist

net: Update address for MediaTek ethernet driver in MAINTAINERS

Update maintainers for MediaTek ethernet driver with Mark Lee.
He is familiar with MediaTek mt762x series ethernet devices and
will keep following maintenance from the vendor side.

Signed-off-by: Sean Wang <[email protected]>
Signed-off-by: Mark Lee <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux

Pull arm64 fixes from Will Deacon:
 "The main thing here is a long-awaited workaround for a CPU erratum on
  ThunderX2 which we have developed in conjunction with engineers from
  Cavium/Marvell.

  At the moment, the workaround is unconditionally enabled for affected
  CPUs at runtime but we may add a command-line option to disable it in
  future if performance numbers show up indicating a significant cost
  for real workloads.

  Summary:

   - Work around Cavium/Marvell ThunderX2 erratum #219

   - Fix regression in mlock() ABI caused by sign-extension of TTBR1 addresses

   - More fixes to the spurious kernel fault detection logic

   - Fix pathological preemption race when enabling some CPU features at boot

   - Drop broken kcore macros in favour of generic implementations

   - Fix userspace view of ID_AA64ZFR0_EL1 when SVE is disabled

   - Avoid NULL dereference on allocation failure during hibernation"

* tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux:
  arm64: tags: Preserve tags for addresses translated via TTBR1
  arm64: mm: fix inverted PAR_EL1.F check
  arm64: sysreg: fix incorrect definition of SYS_PAR_EL1_F
  arm64: entry.S: Do not preempt from IRQ before all cpufeatures are enabled
  arm64: hibernate: check pgd table allocation
  arm64: cpufeature: Treat ID_AA64ZFR0_EL1 as RAZ when SVE is not enabled
  arm64: Fix kcore macros after 52-bit virtual addressing fallout
  arm64: Allow CAVIUM_TX2_ERRATUM_219 to be selected
  arm64: Avoid Cavium TX2 erratum 219 when switching TTBR
  arm64: Enable workaround for Cavium TX2 erratum 219 when running SMT
  arm64: KVM: Trap VM ops when ARM64_WORKAROUND_CAVIUM_TX2_219_TVM is set

ipv4: fix race condition between route lookup and invalidation

Jesse and Ido reported the following race condition:
<CPU A, t0> - Received packet A is forwarded and cached dst entry is
taken from the nexthop ('nhc->nhc_rth_input'). Calls skb_dst_set()

<t1> - Given Jesse has busy routers ("ingesting full BGP routing tables
from multiple ISPs"), route is added / deleted and rt_cache_flush() is
called

<CPU B, t2> - Received packet B tries to use the same cached dst entry
from t0, but rt_cache_valid() is no longer true and it is replaced in
rt_cache_route() by the newer one. This calls dst_dev_put() on the
original dst entry which assigns the blackhole netdev to 'dst->dev'

<CPU A, t3> - dst_input(skb) is called on packet A and it is dropped due
to 'dst->dev' being the blackhole netdev

There are 2 issues in the v4 routing code:
1. A per-netns counter is used to do the validation of the route. That
means whenever a route is changed in the netns, users of all routes in
the netns needs to redo lookup. v6 has an implementation of only
updating fn_sernum for routes that are affected.
2. When rt_cache_valid() returns false, rt_cache_route() is called to
throw away the current cache, and create a new one. This seems
unnecessary because as long as this route does not change, the route
cache does not need to be recreated.

To fully solve the above 2 issues, it probably needs quite some code
changes and requires careful testing, and does not suite for net branch.

So this patch only tries to add the deleted cached rt into the uncached
list, so user could still be able to use it to receive packets until
it's done.

Fixes: 95c47f9cf5e0 ("ipv4: call dst_dev_put() properly")
Signed-off-by: Wei Wang <[email protected]>
Reported-by: Ido Schimmel <[email protected]>
Reported-by: Jesse Hathaway <[email protected]>
Tested-by: Jesse Hathaway <[email protected]>
Acked-by: Martin KaFai Lau <[email protected]>
Cc: David Ahern <[email protected]>
Reviewed-by: Ido Schimmel <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge tag 'xtensa-20191017' of git://github.com/jcmvbkbc/linux-xtensa

Pull Xtensa fixes from Max Filippov:

 - fix {get,put}_user() for 64bit values

 - fix warning about static EXPORT_SYMBOL from modpost

 - fix PCI IO ports mapping for the virt board

 - fix pasto in change_bit for exclusive access option

* tag 'xtensa-20191017' of git://github.com/jcmvbkbc/linux-xtensa:
  xtensa: fix change_bit in exclusive access option
  xtensa: virt: fix PCI IO ports mapping
  xtensa: drop EXPORT_SYMBOL for outs*/ins*
  xtensa: fix type conversion in __get_user_[no]check
  xtensa: clean up assembly arguments in uaccess macros
  xtensa: fix {get,put}_user() for 64bit values

ipv4: Return -ENETUNREACH if we can't create route but saddr is valid

...instead of -EINVAL. An issue was found with older kernel versions
while unplugging a NFS client with pending RPCs, and the wrong error
code here prevented it from recovering once link is back up with a
configured address.

Incidentally, this is not an issue anymore since commit 4f8943f80883
("SUNRPC: Replace direct task wakeups from softirq context"), included
in 5.2-rc7, had the effect of decoupling the forwarding of this error
by using SO_ERROR in xs_wake_error(), as pointed out by Benjamin
Coddington.

To the best of my knowledge, this isn't currently causing any further
issue, but the error code doesn't look appropriate anyway, and we
might hit this in other paths as well.

In detail, as analysed by Gonzalo Siero, once the route is deleted
because the interface is down, and can't be resolved and we return
-EINVAL here, this ends up, courtesy of inet_sk_rebuild_header(),
as the socket error seen by tcp_write_err(), called by
tcp_retransmit_timer().

In turn, tcp_write_err() indirectly calls xs_error_report(), which
wakes up the RPC pending tasks with a status of -EINVAL. This is then
seen by call_status() in the SUN RPC implementation, which aborts the
RPC call calling rpc_exit(), instead of handling this as a
potentially temporary condition, i.e. as a timeout.

Return -EINVAL only if the input parameters passed to
ip_route_output_key_hash_rcu() are actually invalid (this is the case
if the specified source address is multicast, limited broadcast or
all zeroes), but return -ENETUNREACH in all cases where, at the given
moment, the given source address doesn't allow resolving the route.

While at it, drop the initialisation of err to -ENETUNREACH, which
was added to __ip_route_output_key() back then by commit
0315e3827048 ("net: Fix behaviour of unreachable, blackhole and
prohibit routes"), but actually had no effect, as it was, and is,
overwritten by the fib_lookup() return code assignment, and anyway
ignored in all other branches, including the if (fl4->saddr) one:
I find this rather confusing, as it would look like -ENETUNREACH is
the "default" error, while that statement has no effect.

Also note that after commit fc75fc8339e7 ("ipv4: dont create routes
on down devices"), we would get -ENETUNREACH if the device is down,
but -EINVAL if the source address is specified and we can't resolve
the route, and this appears to be rather inconsistent.

Reported-by: Stefan Walter <[email protected]>
Analysed-by: Benjamin Coddington <[email protected]>
Analysed-by: Gonzalo Siero <[email protected]>
Signed-off-by: Stefano Brivio <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

net: phy: micrel: Update KSZ87xx PHY name

The KSZ8795 PHY ID is in fact used by KSZ8794/KSZ8795/KSZ8765 switches.
Update the PHY ID and name to reflect that, as this family of switches
is commonly refered to as KSZ87xx

Signed-off-by: Marek Vasut <[email protected]>
Cc: Andrew Lunn <[email protected]>
Cc: David S. Miller <[email protected]>
Cc: Florian Fainelli <[email protected]>
Cc: George McCollister <[email protected]>
Cc: Heiner Kallweit <[email protected]>
Cc: Sean Nyekjaer <[email protected]>
Cc: Tristram Ha <[email protected]>
Cc: Woojung Huh <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

net: phy: micrel: Discern KSZ8051 and KSZ8795 PHYs

The KSZ8051 PHY and the KSZ8794/KSZ8795/KSZ8765 switch share exactly the
same PHY ID. Since KSZ8051 is higher in the ksphy_driver[] list of PHYs
in the micrel PHY driver, it is used even with the KSZ87xx switch. This
is wrong, since the KSZ8051 configures registers of the PHY which are
not present on the simplified KSZ87xx switch PHYs and misconfigures
other registers of the KSZ87xx switch PHYs.

Fortunatelly, it is possible to tell apart the KSZ8051 PHY from the
KSZ87xx switch by checking the Basic Status register Bit 0, which is
read-only and indicates presence of the Extended Capability Registers.
The KSZ8051 PHY has those registers while the KSZ87xx switch does not.

This patch implements simple check for the presence of this bit for
both the KSZ8051 PHY and KSZ87xx switch, to let both use the correct
PHY driver instance.

Fixes: 9d162ed69f51 ("net: phy: micrel: add support for KSZ8795")
Signed-off-by: Marek Vasut <[email protected]>
Cc: Andrew Lunn <[email protected]>
Cc: David S. Miller <[email protected]>
Cc: Florian Fainelli <[email protected]>
Cc: George McCollister <[email protected]>
Cc: Heiner Kallweit <[email protected]>
Cc: Sean Nyekjaer <[email protected]>
Cc: Tristram Ha <[email protected]>
Cc: Woojung Huh <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

io_uring: fix logic error in io_timeout

If ctx->cached_sq_head < nxt_sq_head, we should add UINT_MAX to tmp, not
tmp_nxt.

Fixes: 5da0fb1ab34c ("io_uring: consider the overflow of sequence for timeout req")
Signed-off-by: yangerkun <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>

io_uring: fix up O_NONBLOCK handling for sockets

We've got two issues with the non-regular file handling for non-blocking
IO:

1) We don't want to re-do a short read in full for a non-regular file,
   as we can't just read the data again.
2) For non-regular files that don't support non-blocking IO attempts,
   we need to punt to async context even if the file is opened as
   non-blocking. Otherwise the caller always gets -EAGAIN.

Add two new request flags to handle these cases. One is just a cache
of the inode S_ISREG() status, the other tells io_uring that we always
need to punt this request to async context, even if REQ_F_NOWAIT is set.

Cc: [email protected]
Reported-by: Hrvoje Zeba <[email protected]>
Tested-by: Hrvoje Zeba <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>

Merge tag 'xfs-5.4-fixes-4' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux

Pull xfs fix from Darrick Wong:
 "The single fix converts the seconds field in the recently added XFS
  bulkstat structure to a signed 64-bit quantity.

  The structure layout doesn't change and so far there are no users of
  the ioctl to break because we only publish xfs ioctl interfaces
  through the XFS userspace development libraries, and we're still
  working on a 5.3 release"

* tag 'xfs-5.4-fixes-4' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux:
  xfs: change the seconds fields in xfs_bulkstat to signed

Merge tag 'drm-fixes-2019-10-18' of git://anongit.freedesktop.org/drm/drm

Pull drm fixes from Dave Airlie:
 "This is this weeks fixes for drm.

  The dma-resv one is probably the more important one a fair few people
  have reported it, besides that it's a couple of panfrost, a few i915
  and a few amdgpu fixes.

  One radeon patch to fix some ppc64 related issues caused an x86
  regression so is getting reverted for now.

  Summary:

  dma-resv:
   - shared fences for lima/panfrost

  ttm:
   - prefault regression fix
   - lifetime fix

  panfrost:
   - stopped job timeout fix
   - missing register values

  amdgpu:
   - smu7 powerplay fix
   - bail earlier for cik/si detection
   - navi SDMA fix

  radeon:
   - revert a ppc64 shutdown fix that broke x86

  i915:
   - VBT information handling fix
   - Circular locking fix
   - preemption vs resubmission virtual requests fix"

* tag 'drm-fixes-2019-10-18' of git://anongit.freedesktop.org/drm/drm:
  drm/i915: Fixup preempt-to-busy vs resubmission of a virtual request
  drm/i915/userptr: Never allow userptr into the mappable GGTT
  drm/i915: Favor last VBT child device with conflicting AUX ch/DDC pin
  drm/i915/execlists: Refactor -EIO markup of hung requests
  drm/panfrost: Handle resetting on timeout better
  drm/panfrost: Add missing GPU feature registers
  drm/ttm: fix handling in ttm_bo_add_mem_to_lru
  drm/ttm: Restore ttm prefaulting
  drm/ttm: fix busy reference in ttm_mem_evict_first
  drm/amdgpu/sdma5: fix mask value of POLL_REGMEM packet for pipe sync
  drm/amdgpu: Bail earlier when amdgpu.cik_/si_support is not set to 1
  Revert "drm/radeon: Fix EEH during kexec"
  drm/msm/dsi: Implement reset correctly
  dma-buf/resv: fix exclusive fence get
  drm/edid: Add 6 bpc quirk for SDC panel in Lenovo G50
  drm/tiny: Kconfig: Remove always-y THERMAL dep. from TINYDRM_REPAPER
  drm/amdgpu/powerplay: fix typo in mvdd table setup

Merge branch 'errata/tx2-219' into for-next/fixes

Workaround for Cavium/Marvell ThunderX2 erratum #219.

* errata/tx2-219:
  arm64: Allow CAVIUM_TX2_ERRATUM_219 to be selected
  arm64: Avoid Cavium TX2 erratum 219 when switching TTBR
  arm64: Enable workaround for Cavium TX2 erratum 219 when running SMT
  arm64: KVM: Trap VM ops when ARM64_WORKAROUND_CAVIUM_TX2_219_TVM is set

Merge tag 'drm-misc-fixes-2019-10-17' of git://anongit.freedesktop.org/drm/drm-misc into drm-fixes

-dma-resv: Change shared_count to post-increment to fix lima crash (Qiang)
-ttm: A couple fixes related to lifetime and restore prefault behavior
 (Christian & Thomas)
-panfrost: Fill in missing feature reg values and fix stoppedjob timeouts
 (Steven)

Cc: Qiang Yu <[email protected]>
Cc: Thomas Hellstrom <[email protected]>
Cc: Christian König <[email protected]>
Cc: Steven Price <[email protected]>
Signed-off-by: Dave Airlie <[email protected]>
From: Sean Paul <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/20191017203419.GA142909@art_vandelay

Merge tag 'drm-fixes-5.4-2019-10-16' of git://people.freedesktop.org/~agd5f/linux into drm-fixes

drm-fixes-5.4-2019-10-16:

amdgpu:
- Powerplay fix for SMU7 parts
- Bail earlier when cik/si support is not set to 1
- Fix an SDMA issue on navi

radeon:
- revert a PPC fix which broken x86

Signed-off-by: Dave Airlie <[email protected]>
From: Alex Deucher <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]

Merge tag 'drm-intel-fixes-2019-10-17' of git://anongit.freedesktop.org/drm/drm-intel into drm-fixes

- Display fix on handling VBT information.
- Important circular locking fix
- Fix for preemption vs resubmission on virtual requests
  - and a prep patch to make this last one to apply cleanly

Signed-off-by: Dave Airlie <[email protected]>
From: Rodrigo Vivi <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]

net: dsa: microchip: Add shared regmap mutex

The KSZ driver uses one regmap per register width (8/16/32), each with
it's own lock, but accessing the same set of registers. In theory, it
is possible to create a race condition between these regmaps, although
the underlying bus (SPI or I2C) locking should assure nothing bad will
really happen and the accesses would be correct.

To make the driver do the right thing, add one single shared mutex for
all the regmaps used by the driver instead. This assures that even if
some future hardware is on a bus which does not serialize the accesses
the same way SPI or I2C does, nothing bad will happen.

Note that the status_mutex was unused and only initied, hence it was
renamed and repurposed as the regmap mutex.

Signed-off-by: Marek Vasut <[email protected]>
Cc: Andrew Lunn <[email protected]>
Cc: David S. Miller <[email protected]>
Cc: Florian Fainelli <[email protected]>
Cc: George McCollister <[email protected]>
Cc: Tristram Ha <[email protected]>
Cc: Woojung Huh <[email protected]>
Reviewed-by: Andrew Lunn <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

net: dsa: microchip: Do not reinit mutexes on KSZ87xx

The KSZ87xx driver calls mutex_init() on mutexes already inited in
ksz_common.c ksz_switch_register(). Do not do it twice, drop the
reinitialization.

Signed-off-by: Marek Vasut <[email protected]>
Cc: Andrew Lunn <[email protected]>
Cc: David S. Miller <[email protected]>
Cc: Florian Fainelli <[email protected]>
Cc: George McCollister <[email protected]>
Cc: Tristram Ha <[email protected]>
Cc: Woojung Huh <[email protected]>
Reviewed-by: Andrew Lunn <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

net: stmmac: fix argument to stmmac_pcs_ctrl_ane()

The stmmac_pcs_ctrl_ane() expects a register address as
argument 1, but for some reason the mac_device_info is
being passed.

Fix the warning (and possible bug) from sparse:

drivers/net/ethernet/stmicro/stmmac/stmmac_main.c:2613:17: warning: incorrect type in argument 1 (different address spaces)
drivers/net/ethernet/stmicro/stmmac/stmmac_main.c:2613:17:    expected void [noderef] <asn:2> *ioaddr
drivers/net/ethernet/stmicro/stmmac/stmmac_main.c:2613:17:    got struct mac_device_info *hw

Signed-off-by: Ben Dooks <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge branch 'dpaa2-eth-misc-fixes'

Ioana Ciornei says:

====================
dpaa2-eth: misc fixes

This patch set adds a couple of fixes around updating configuration on MAC
change.  Depending on when MC connects the DPNI to a MAC, both the MAC
address and TX FQIDs should be updated everytime there is a change in
configuration.

Changes in v2:
 - used reverse christmas tree ordering in patch 2/2
Changes in v3:
 - add a missing new line
 - go back to FQ based enqueueing after a transient error
====================

Signed-off-by: David S. Miller <[email protected]>

dpaa2-eth: Fix TX FQID values

Depending on when MC connects the DPNI to a MAC, Tx FQIDs may
not be available during probe time.

Read the FQIDs each time the link goes up to avoid using invalid
values. In case an error occurs or an invalid value is retrieved,
fall back to QDID-based enqueueing.

Fixes: 1fa0f68c9255 ("dpaa2-eth: Use FQ-based DPIO enqueue API")
Signed-off-by: Ioana Radulescu <[email protected]>
Signed-off-by: Ioana Ciornei <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

dpaa2-eth: add irq for the dpmac connect/disconnect event

Add IRQ for the DPNI endpoint change event, resolving the issue
when a dynamically created DPNI gets a randomly generated hw address
when the endpoint is a DPMAC object.

Signed-off-by: Florin Chiculita <[email protected]>
Signed-off-by: Ioana Ciornei <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

usb: hso: obey DMA rules in tiocmget

The serial state information must not be embedded into another
data structure, as this interferes with cache handling for DMA
on architectures without cache coherence..
That would result in data corruption on some architectures
Allocating it separately.

v2: fix syntax error

Signed-off-by: Oliver Neukum <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input

Pull input fixes from Dmitry Torokhov:
 "The main change is that we are reverting blanket enablement of SMBus
  mode for devices with Elan touchpads that report BIOS release date as
  2018+ because there are older boxes with updated BIOSes that still do
  not work well in SMbus mode.

  We will have to establish whitelist for SMBus mode it looks like"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input:
  Revert "Input: elantech - enable SMBus on new (2018+) systems"
  Input: synaptics-rmi4 - avoid processing unknown IRQs
  Input: soc_button_array - partial revert of support for newer surface devices
  Input: goodix - add support for 9-bytes reports
  Input: da9063 - fix capability and drop KEY_SLEEP

coccinelle: api/devm_platform_ioremap_resource: remove useless script

While it is useful for new drivers to use devm_platform_ioremap_resource,
this script is currently used to spam maintainers, often updating very
old drivers.  The net benefit is the removal of 2 lines of code in the
driver but the review load for the maintainers is huge.  As of now, more
that 560 patches have been sent, some of them obviously broken, as in:

 https://lore.kernel.org/lkml/9bbcce19c777583815c92ce3c2ff2586@www.loen.fr/

Remove the script to reduce the spam.

Signed-off-by: Alexandre Belloni <[email protected]>
Acked-by: Julia Lawall <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

ALSA: hda - Force runtime PM on Nvidia HDMI codecs

Przemysław Kopa reports that since commit b516ea586d71 ("PCI: Enable
NVIDIA HDA controllers"), the discrete GPU Nvidia GeForce GT 540M on his
2011 Samsung laptop refuses to runtime suspend, resulting in a power
regression and excessive heat.

Rivera Valdez witnesses the same issue with a GeForce GT 525M (GF108M)
of the same era, as does another Arch Linux user named "R0AR" with a
more recent GeForce GTX 1050 Ti (GP107M).

The commit exposes the discrete GPU's HDA controller and all four codecs
on the controller do not set the CLKSTOP and EPSS bits in the Supported
Power States Response.  They also do not set the PS-ClkStopOk bit in the
Get Power State Response.  hda_codec_runtime_suspend() therefore does
not call snd_hdac_codec_link_down(), which prevents each codec and the
PCI device from runtime suspending.

The same issue is present on some AMD discrete GPUs and we addressed it
by forcing runtime PM despite the bits not being set, see commit
57cb54e53bdd ("ALSA: hda - Force to link down at runtime suspend on
ATI/AMD HDMI").

Do the same for Nvidia HDMI codecs.

Fixes: b516ea586d71 ("PCI: Enable NVIDIA HDA controllers")
Link: https://bbs.archlinux.org/viewtopic.php?pid=1865512
Link: https://bugs.freedesktop.org/show_bug.cgi?id=75985#c81
Reported-by: Przemysław Kopa <[email protected]>
Reported-by: Rivera Valdez <[email protected]>
Signed-off-by: Lukas Wunner <[email protected]>
Cc: Daniel Drake <[email protected]>
Cc: [email protected] # v5.3+
Link: https://lore.kernel.org/r/3086bc75135c1e3567c5bc4f3cc4ff5cbf7a56c2.1571324194.git.lukas@wunner.de
Signed-off-by: Takashi Iwai <[email protected]>

Merge tag 'platform-drivers-x86-v5.4-3' of git://git.infradead.org/linux-platform-drivers-x86

Pull x86 platform driver fixes from Andy Shevchenko:

 - Users of Intel P-Unit IPC driver might be surprised by harmless
   warning. Thus, switch to API which doesn't issue a warning at all.

 - I²C multi-instantiate driver continues to add slave devices even when
   IRQ resource is not found. For devices in the market IRQ resource is
   mandatory, so, fail the ->probe() of the parent driver to avoid
   slaves being probed.

 - Avoid compiler warning due to unused variable in Classmate laptop
   driver.

* tag 'platform-drivers-x86-v5.4-3' of git://git.infradead.org/linux-platform-drivers-x86:
  platform/x86: i2c-multi-instantiate: Fail the probe if no IRQ provided
  platform/x86: intel_punit_ipc: Avoid error message when retrieving IRQ
  platform/x86: classmate-laptop: remove unused variable

dm cache: fix bugs when a GFP_NOWAIT allocation fails

GFP_NOWAIT allocation can fail anytime - it doesn't wait for memory being
available and it fails if the mempool is exhausted and there is not enough
memory.

If we go down this path:
  map_bio -> mg_start -> alloc_migration -> mempool_alloc(GFP_NOWAIT)
we can see that map_bio() doesn't check the return value of mg_start(),
and the bio is leaked.

If we go down this path:
  map_bio -> mg_start -> mg_lock_writes -> alloc_prison_cell ->
  dm_bio_prison_alloc_cell_v2 -> mempool_alloc(GFP_NOWAIT) ->
  mg_lock_writes -> mg_complete
the bio is ended with an error - it is unacceptable because it could
cause filesystem corruption if the machine ran out of memory
temporarily.

Change GFP_NOWAIT to GFP_NOIO, so that the mempool code will properly
wait until memory becomes available. mempool_alloc with GFP_NOIO can't
fail, so remove the code paths that deal with allocation failure.

Cc: [email protected]
Signed-off-by: Mikulas Patocka <[email protected]>
Signed-off-by: Mike Snitzer <[email protected]>

Merge tag 'gpio-v5.4-3' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-gpio

Pull GPIO fixes from Linus Walleij:
 "The fixes pertain to a problem with initializing the Intel GPIO
  irqchips when adding gpiochips.

  Andy fixed it up elegantly by adding a hardware initialization
  callback to the struct gpio_irq_chip so let's use this. Tested and
  verified on the target hardware"

* tag 'gpio-v5.4-3' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-gpio:
  gpio: lynxpoint: set default handler to be handle_bad_irq()
  gpio: merrifield: Move hardware initialization to callback
  gpio: lynxpoint: Move hardware initialization to callback
  gpio: intel-mid: Move hardware initialization to callback
  gpiolib: Initialize the hardware with a callback
  gpio: merrifield: Restore use of irq_base

ALSA: hda/realtek - Enable headset mic on Asus MJ401TA

On Asus MJ401TA (with Realtek ALC256), the headset mic is connected to
pin 0x19, with default configuration value 0x411111f0 (indicating no
physical connection).

Enable this by quirking the pin. Mic jack detection was also tested and
found to be working.

This enables use of the headset mic on this product.

Signed-off-by: Daniel Drake <[email protected]>
Cc: <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Takashi Iwai <[email protected]>

ALSA: usb-audio: Disable quirks for BOSS Katana amplifiers

BOSS Katana amplifiers cannot be used for recording or playback if quirks
are applied

BugLink: https://bugzilla.kernel.org/show_bug.cgi?id=195223
Signed-off-by: Szabolcs Szőke <[email protected]>
Cc: <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Takashi Iwai <[email protected]>

net: stmmac: disable/enable ptp_ref_clk in suspend/resume flow

disable ptp_ref_clk in suspend flow, and enable it in resume flow.

Fixes: f573c0b9c4e0 ("stmmac: move stmmac_clk, pclk, clk_ptp_ref and stmmac_rst to platform structure")
Signed-off-by: Biao Huang <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

net: phy: Fix "link partner" information disappear issue

Some drivers just call phy_ethtool_ksettings_set() to set the
links, for those phy drivers that use genphy_read_status(), if
autoneg is on, and the link is up, than execute "ethtool -s
ethx autoneg on" will cause "link partner" information disappear.

The call trace is phy_ethtool_ksettings_set()->phy_start_aneg()
->linkmode_zero(phydev->lp_advertising)->genphy_read_status(),
the link didn't change, so genphy_read_status() just return, and
phydev->lp_advertising is zero now.

This patch moves the clear operation of lp_advertising from
phy_start_aneg() to genphy_read_lpa()/genphy_c45_read_lpa(), and
if autoneg on and autoneg not complete, just clear what the
generic functions care about.

Fixes: 88d6272acaaa ("net: phy: avoid unneeded MDIO reads in genphy_read_status")
Signed-off-by: Yonglong Liu <[email protected]>
Reviewed-by: Heiner Kallweit <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

rxrpc: use rcu protection while reading sk->sk_user_data

We need to extend the rcu_read_lock() section in rxrpc_error_report()
and use rcu_dereference_sk_user_data() instead of plain access
to sk->sk_user_data to make sure all rules are respected.

The compiler wont reload sk->sk_user_data at will, and RCU rules
prevent memory beeing freed too soon.

Fixes: f0308fb07080 ("rxrpc: Fix possible NULL pointer access in ICMP handling")
Fixes: 17926a79320a ("[AF_RXRPC]: Provide secure RxRPC sockets for use by userspace and kernel both")
Signed-off-by: Eric Dumazet <[email protected]>
Cc: David Howells <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

drm/i915: Fixup preempt-to-busy vs resubmission of a virtual request

As preempt-to-busy leaves the request on the HW as the resubmission is
processed, that request may complete in the background and even cause a
second virtual request to enter queue. This second virtual request
breaks our "single request in the virtual pipeline" assumptions.
Furthermore, as the virtual request may be completed and retired, we
lose the reference the virtual engine assumes is held. Normally, just
removing the request from the scheduler queue removes it from the
engine, but the virtual engine keeps track of its singleton request via
its ve->request. This pointer needs protecting with a reference.

v2: Drop unnecessary motion of rq->engine = owner

Fixes: 22b7a426bbe1 ("drm/i915/execlists: Preempt-to-busy")
Signed-off-by: Chris Wilson <[email protected]>
Cc: Mika Kuoppala <[email protected]>
Cc: Tvrtko Ursulin <[email protected]>
Reviewed-by: Tvrtko Ursulin <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]
(cherry picked from commit b647c7df01b75761b4c0b1cb6f4841088c0b1121)
Signed-off-by: Rodrigo Vivi <[email protected]>

drm/i915/userptr: Never allow userptr into the mappable GGTT

Daniel Vetter uncovered a nasty cycle in using the mmu-notifiers to
invalidate userptr objects which also happen to be pulled into GGTT
mmaps. That is when we unbind the userptr object (on mmu invalidation),
we revoke all CPU mmaps, which may then recurse into mmu invalidation.

We looked for ways of breaking the cycle, but the revocation on
invalidation is required and cannot be avoided. The only solution we
could see was to not allow such GGTT bindings of userptr objects in the
first place. In practice, no one really wants to use a GGTT mmapping of
a CPU pointer...

Just before Daniel's explosive lockdep patches land in v5.4-rc1, we got
a genuine blip from CI:

<4>[  246.793958] ======================================================
<4>[  246.793972] WARNING: possible circular locking dependency detected
<4>[  246.793989] 5.3.0-gbd6c56f50d15-drmtip_372+ #1 Tainted: G     U
<4>[  246.794003] ------------------------------------------------------
<4>[  246.794017] kswapd0/145 is trying to acquire lock:
<4>[  246.794030] 000000003f565be6 (&dev->struct_mutex/1){+.+.}, at: userptr_mn_invalidate_range_start+0x18f/0x220 [i915]
<4>[  246.794250]
                  but task is already holding lock:
<4>[  246.794263] 000000001799cef9 (&anon_vma->rwsem){++++}, at: page_lock_anon_vma_read+0xe6/0x2a0
<4>[  246.794291]
                  which lock already depends on the new lock.

<4>[  246.794307]
                  the existing dependency chain (in reverse order) is:
<4>[  246.794322]
                  -> #3 (&anon_vma->rwsem){++++}:
<4>[  246.794344]        down_write+0x33/0x70
<4>[  246.794357]        __vma_adjust+0x3d9/0x7b0
<4>[  246.794370]        __split_vma+0x16a/0x180
<4>[  246.794385]        mprotect_fixup+0x2a5/0x320
<4>[  246.794399]        do_mprotect_pkey+0x208/0x2e0
<4>[  246.794413]        __x64_sys_mprotect+0x16/0x20
<4>[  246.794429]        do_syscall_64+0x55/0x1c0
<4>[  246.794443]        entry_SYSCALL_64_after_hwframe+0x49/0xbe
<4>[  246.794456]
                  -> #2 (&mapping->i_mmap_rwsem){++++}:
<4>[  246.794478]        down_write+0x33/0x70
<4>[  246.794493]        unmap_mapping_pages+0x48/0x130
<4>[  246.794519]        i915_vma_revoke_mmap+0x81/0x1b0 [i915]
<4>[  246.794519]        i915_vma_unbind+0x11d/0x4a0 [i915]
<4>[  246.794519]        i915_vma_destroy+0x31/0x300 [i915]
<4>[  246.794519]        __i915_gem_free_objects+0xb8/0x4b0 [i915]
<4>[  246.794519]        drm_file_free.part.0+0x1e6/0x290
<4>[  246.794519]        drm_release+0xa6/0xe0
<4>[  246.794519]        __fput+0xc2/0x250
<4>[  246.794519]        task_work_run+0x82/0xb0
<4>[  246.794519]        do_exit+0x35b/0xdb0
<4>[  246.794519]        do_group_exit+0x34/0xb0
<4>[  246.794519]        __x64_sys_exit_group+0xf/0x10
<4>[  246.794519]        do_syscall_64+0x55/0x1c0
<4>[  246.794519]        entry_SYSCALL_64_after_hwframe+0x49/0xbe
<4>[  246.794519]
                  -> #1 (&vm->mutex){+.+.}:
<4>[  246.794519]        i915_gem_shrinker_taints_mutex+0x6d/0xe0 [i915]
<4>[  246.794519]        i915_address_space_init+0x9f/0x160 [i915]
<4>[  246.794519]        i915_ggtt_init_hw+0x55/0x170 [i915]
<4>[  246.794519]        i915_driver_probe+0xc9f/0x1620 [i915]
<4>[  246.794519]        i915_pci_probe+0x43/0x1b0 [i915]
<4>[  246.794519]        pci_device_probe+0x9e/0x120
<4>[  246.794519]        really_probe+0xea/0x3d0
<4>[  246.794519]        driver_probe_device+0x10b/0x120
<4>[  246.794519]        device_driver_attach+0x4a/0x50
<4>[  246.794519]        __driver_attach+0x97/0x130
<4>[  246.794519]        bus_for_each_dev+0x74/0xc0
<4>[  246.794519]        bus_add_driver+0x13f/0x210
<4>[  246.794519]        driver_register+0x56/0xe0
<4>[  246.794519]        do_one_initcall+0x58/0x300
<4>[  246.794519]        do_init_module+0x56/0x1f6
<4>[  246.794519]        load_module+0x25bd/0x2a40
<4>[  246.794519]        __se_sys_finit_module+0xd3/0xf0
<4>[  246.794519]        do_syscall_64+0x55/0x1c0
<4>[  246.794519]        entry_SYSCALL_64_after_hwframe+0x49/0xbe
<4>[  246.794519]
                  -> #0 (&dev->struct_mutex/1){+.+.}:
<4>[  246.794519]        __lock_acquire+0x15d8/0x1e90
<4>[  246.794519]        lock_acquire+0xa6/0x1c0
<4>[  246.794519]        __mutex_lock+0x9d/0x9b0
<4>[  246.794519]        userptr_mn_invalidate_range_start+0x18f/0x220 [i915]
<4>[  246.794519]        __mmu_notifier_invalidate_range_start+0x85/0x110
<4>[  246.794519]        try_to_unmap_one+0x76b/0x860
<4>[  246.794519]        rmap_walk_anon+0x104/0x280
<4>[  246.794519]        try_to_unmap+0xc0/0xf0
<4>[  246.794519]        shrink_page_list+0x561/0xc10
<4>[  246.794519]        shrink_inactive_list+0x220/0x440
<4>[  246.794519]        shrink_node_memcg+0x36e/0x740
<4>[  246.794519]        shrink_node+0xcb/0x490
<4>[  246.794519]        balance_pgdat+0x241/0x580
<4>[  246.794519]        kswapd+0x16c/0x530
<4>[  246.794519]        kthread+0x119/0x130
<4>[  246.794519]        ret_from_fork+0x24/0x50
<4>[  246.794519]
                  other info that might help us debug this:

<4>[  246.794519] Chain exists of:
                    &dev->struct_mutex/1 --> &mapping->i_mmap_rwsem --> &anon_vma->rwsem

<4>[  246.794519]  Possible unsafe locking scenario:

<4>[  246.794519]        CPU0                    CPU1
<4>[  246.794519]        ----                    ----
<4>[  246.794519]   lock(&anon_vma->rwsem);
<4>[  246.794519]                                lock(&mapping->i_mmap_rwsem);
<4>[  246.794519]                                lock(&anon_vma->rwsem);
<4>[  246.794519]   lock(&dev->struct_mutex/1);
<4>[  246.794519]
                   *** DEADLOCK ***

v2: Say no to mmap_ioctl

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111744
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111870
Signed-off-by: Chris Wilson <[email protected]>
Cc: Tvrtko Ursulin <[email protected]>
Cc: Daniel Vetter <[email protected]>
Cc: [email protected]
Reviewed-by: Tvrtko Ursulin <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]
(cherry picked from commit a4311745bba9763e3c965643d4531bd5765b0513)
Signed-off-by: Rodrigo Vivi <[email protected]>

drm/i915: Favor last VBT child device with conflicting AUX ch/DDC pin

The first come first served apporoach to handling the VBT
child device AUX ch conflicts has backfired. We have machines
in the wild where the VBT specifies both port A eDP and
port E DP (in that order) with port E being the real one.

So let's try to flip the preference around and let the last
child device win once again.

Cc: [email protected]
Cc: Jani Nikula <[email protected]>
Tested-by: Masami Ichikawa <[email protected]>
Tested-by: Torsten <[email protected]>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111966
Fixes: 36a0f92020dc ("drm/i915/bios: make child device order the priority order")
Signed-off-by: Ville Syrjälä <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]
Acked-by: Jani Nikula <[email protected]>
(cherry picked from commit 41e35ffb380bde1379e4030bb5b2ac824d5139cf)
Signed-off-by: Rodrigo Vivi <[email protected]>

drm/i915/execlists: Refactor -EIO markup of hung requests

Pull setting -EIO on the hung requests into its own utility function.
Having allowed ourselves to short-circuit submission of completed
requests, we can now do the mark_eio() prior to submission and avoid
some redundant operations.

Signed-off-by: Chris Wilson <[email protected]>
Reviewed-by: Tvrtko Ursulin <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]
(cherry picked from commit 0d7cf7bc15e75bf79f2f65d61d19f896609f816a)
Signed-off-by: Rodrigo Vivi <[email protected]>

Revert "blackhole_netdev: fix syzkaller reported issue"

This reverts commit b0818f80c8c1bc215bba276bd61c216014fab23b.

Started seeing weird behavior after this patch especially in
the IPv6 code path. Haven't root caused it, but since this was
applied to net branch, taking a precautionary measure to revert
it and look / analyze those failures

Revert this now and I'll send a better fix after analysing / fixing
the weirdness observed.

CC: Eric Dumazet <[email protected]>
CC: Wei Wang <[email protected]>
CC: David S. Miller <[email protected]>
Signed-off-by: Mahesh Bandewar <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

arm64: tags: Preserve tags for addresses translated via TTBR1

Sign-extending TTBR1 addresses when converting to an untagged address
breaks the documented POSIX semantics for mlock() in some obscure error
cases where we end up returning -EINVAL instead of -ENOMEM as a direct
result of rewriting the upper address bits.

Rework the untagged_addr() macro to preserve the upper address bits for
TTBR1 addresses and only clear the tag bits for user addresses. This
matches the behaviour of the 'clear_address_tag' assembly macro, so
rename that and align the implementations at the same time so that they
use the same instruction sequences for the tag manipulation.

Link: https://lore.kernel.org/stable/[email protected]/
Reported-by: Jan Stancek <[email protected]>
Tested-by: Jan Stancek <[email protected]>
Reviewed-by: Catalin Marinas <[email protected]>
Tested-by: Catalin Marinas <[email protected]>
Reviewed-by: Vincenzo Frascino <[email protected]>
Tested-by: Vincenzo Frascino <[email protected]>
Reviewed-by: Andrey Konovalov <[email protected]>
Signed-off-by: Will Deacon <[email protected]>

arm64: mm: fix inverted PAR_EL1.F check

When detecting a spurious EL1 translation fault, we have the CPU retry
the translation using an AT S1E1R instruction, and inspect PAR_EL1 to
determine if the fault was spurious.

When PAR_EL1.F == 0, the AT instruction successfully translated the
address without a fault, which implies the original fault was spurious.
However, in this case we return false and treat the original fault as if
it was not spurious.

Invert the return value so that we treat such a case as spurious.

Cc: Catalin Marinas <[email protected]>
Fixes: 42f91093b043 ("arm64: mm: Ignore spurious translation faults taken from the kernel")
Tested-by: James Morse <[email protected]>
Signed-off-by: Mark Rutland <[email protected]>
Signed-off-by: Will Deacon <[email protected]>

arm64: sysreg: fix incorrect definition of SYS_PAR_EL1_F

The 'F' field of the PAR_EL1 register lives in bit 0, not bit 1.
Fix the broken definition in 'sysreg.h'.

Fixes: e8620cff9994 ("arm64: sysreg: Add some field definitions for PAR_EL1")
Reviewed-by: Mark Rutland <[email protected]>
Signed-off-by: Yang Yingliang <[email protected]>
Signed-off-by: Will Deacon <[email protected]>

arm64: entry.S: Do not preempt from IRQ before all cpufeatures are enabled

Preempting from IRQ-return means that the task has its PSTATE saved
on the stack, which will get restored when the task is resumed and does
the actual IRQ return.

However, enabling some CPU features requires modifying the PSTATE. This
means that, if a task was scheduled out during an IRQ-return before all
CPU features are enabled, the task might restore a PSTATE that does not
include the feature enablement changes once scheduled back in.

* Task 1:

PAN == 0 ---|                          |---------------
            |                          |<- return from IRQ, PSTATE.PAN = 0
            | <- IRQ                   |
            +--------+ <- preempt()  +--
                                     ^
                                     |
                                     reschedule Task 1, PSTATE.PAN == 1
* Init:
        --------------------+------------------------
                            ^
                            |
                            enable_cpu_features
                            set PSTATE.PAN on all CPUs

Worse than this, since PSTATE is untouched when task switching is done,
a task missing the new bits in PSTATE might affect another task, if both
do direct calls to schedule() (outside of IRQ/exception contexts).

Fix this by preventing preemption on IRQ-return until features are
enabled on all CPUs.

This way the only PSTATE values that are saved on the stack are from
synchronous exceptions. These are expected to be fatal this early, the
exception is BRK for WARN_ON(), but as this uses do_debug_exception()
which keeps IRQs masked, it shouldn't call schedule().

Signed-off-by: Julien Thierry <[email protected]>
[james: Replaced a really cool hack, with an even simpler static key in C.
 expanded commit message with Julien's cover-letter ascii art]
Signed-off-by: James Morse <[email protected]>
Signed-off-by: Will Deacon <[email protected]>

Merge branch 'md-fixes' of https://git.kernel.org/pub/scm/linux/kernel/git/song/md into for-linus

Pull MD fix from Song.

* 'md-fixes' of https://git.kernel.org/pub/scm/linux/kernel/git/song/md:
  md/raid0: fix warning message for parameter default_layout

md/raid0: fix warning message for parameter default_layout

The message should match the parameter, i.e. raid0.default_layout.

Fixes: c84a1372df92 ("md/raid0: avoid RAID0 data corruption due to layout confusion.")
Cc: NeilBrown <[email protected]>
Reported-by: Ivan Topolsky <[email protected]>
Signed-off-by: Song Liu <[email protected]>

kthread: make __kthread_queue_delayed_work static

The __kthread_queue_delayed_work is not exported so
make it static, to avoid the following sparse warning:

  kernel/kthread.c:869:6: warning: symbol '__kthread_queue_delayed_work' was not declared. Should it be static?

Signed-off-by: Ben Dooks <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

usercopy: Avoid soft lockups in test_check_nonzero_user()

On a machine with a 64K PAGE_SIZE, the nested for loops in
test_check_nonzero_user() can lead to soft lockups, eg:

  watchdog: BUG: soft lockup - CPU#4 stuck for 22s! [modprobe:611]
  Modules linked in: test_user_copy(+) vmx_crypto gf128mul crc32c_vpmsum virtio_balloon ip_tables x_tables autofs4
  CPU: 4 PID: 611 Comm: modprobe Tainted: G             L    5.4.0-rc1-gcc-8.2.0-00001-gf5a1a536fa14-dirty #1151
  ...
  NIP __might_sleep+0x20/0xc0
  LR  __might_fault+0x40/0x60
  Call Trace:
    check_zeroed_user+0x12c/0x200
    test_user_copy_init+0x67c/0x1210 [test_user_copy]
    do_one_initcall+0x60/0x340
    do_init_module+0x7c/0x2f0
    load_module+0x2d94/0x30e0
    __do_sys_finit_module+0xc8/0x150
    system_call+0x5c/0x68

Even with a 4K PAGE_SIZE the test takes multiple seconds. Instead
tweak it to only scan a 1024 byte region, but make it cross the
page boundary.

Fixes: f5a1a536fa14 ("lib: introduce copy_struct_from_user() helper")
Suggested-by: Aleksa Sarai <[email protected]>
Signed-off-by: Michael Ellerman <[email protected]>
Reviewed-by: Aleksa Sarai <[email protected]>
Acked-by: Christian Brauner <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Christian Brauner <[email protected]>

ACPI: processor: Avoid NULL pointer dereferences at init time

If there are neither processor objects nor processor device objects
in the ACPI tables, the per-CPU processors table will not be
initialized and attempting to dereference pointers from there will
cause the kernel to crash.  This happens in acpi_processor_ppc_init()
and acpi_thermal_cpufreq_init() after commit d15ce412737a ("ACPI:
cpufreq: Switch to QoS requests instead of cpufreq notifier")
which didn't add the requisite NULL pointer checks in there.

Add the NULL pointer checks to acpi_processor_ppc_init() and
acpi_thermal_cpufreq_init(), and to the corresponding "exit"
routines.

While at it, drop redundant return instructions from
acpi_processor_ppc_init() and acpi_thermal_cpufreq_init().

Fixes: d15ce412737a ("ACPI: cpufreq: Switch to QoS requests instead of cpufreq notifier")
Reported-by: Srinivas Pandruvada <[email protected]>
Signed-off-by: Rafael J. Wysocki <[email protected]>
Acked-by: Viresh Kumar <[email protected]>

xtensa: fix change_bit in exclusive access option

change_bit implementation for XCHAL_HAVE_EXCLUSIVE case changes all bits
except the one required due to copy-paste error from clear_bit.

Cc: [email protected] # v5.2+
Fixes: f7c34874f04a ("xtensa: add exclusive atomics support")
Signed-off-by: Max Filippov <[email protected]>

RISC-V: fix virtual address overlapped in FIXADDR_START and VMEMMAP_START

This patch fixes the virtual address layout in pgtable.h.  The virtual
address of FIXADDR_START and VMEMMAP_START should not be overlapped.

Fixes: d95f1a542c3d ("RISC-V: Implement sparsemem")
Signed-off-by: Greentime Hu <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
[[email protected]: fixed patch description]
Signed-off-by: Paul Walmsley <[email protected]>

net: usb: sr9800: fix uninitialized local variable

Make sure res does not contain random value if the call to
sr_read_cmd fails for some reason.

Reported-by: [email protected]
Signed-off-by: Valentin Vidic <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

net: bcmgenet: Fix RGMII_MODE_EN value for GENET v1/2/3

The RGMII_MODE_EN bit value was 0 for GENET versions 1 through 3, and
became 6 for GENET v4 and above, account for that difference.

Fixes: aa09677cba42 ("net: bcmgenet: add MDIO routines")
Signed-off-by: Florian Fainelli <[email protected]>
Acked-by: Doug Berger <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

net: stmmac: make tc_flow_parsers static

The tc_flow_parsers is not used outside of the driver, so
make it static to avoid the following sparse warning:

drivers/net/ethernet/stmicro/stmmac/stmmac_tc.c:516:3: warning: symbol 'tc_flow_parsers' was not declared. Should it be static?

Signed-off-by: Ben Dooks <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

davinci_cpdma: make cpdma_chan_split_pool static

The cpdma_chan_split_pool() function is not used outside of
the driver, so make it static to avoid the following sparse
warning:

drivers/net/ethernet/ti/davinci_cpdma.c:725:5: warning: symbol 'cpdma_chan_split_pool' was not declared. Should it be static?

Signed-off-by: Ben Dooks <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

net: i82596: fix dma_alloc_attr for sni_82596

Commit 7f683b920479 ("i825xx: switch to switch to dma_alloc_attrs")
switched dma allocation over to dma_alloc_attr, but didn't convert
the SNI part to request consistent DMA memory. This broke sni_82596
since driver doesn't do dma_cache_sync for performance reasons.
Fix this by using different DMA_ATTRs for lasi_82596 and sni_82596.

Fixes: 7f683b920479 ("i825xx: switch to switch to dma_alloc_attrs")
Signed-off-by: Thomas Bogendoerfer <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

sctp: change sctp_prot .no_autobind with true

syzbot reported a memory leak:

  BUG: memory leak, unreferenced object 0xffff888120b3d380 (size 64):
  backtrace:

    [...] slab_alloc mm/slab.c:3319 [inline]
    [...] kmem_cache_alloc+0x13f/0x2c0 mm/slab.c:3483
    [...] sctp_bucket_create net/sctp/socket.c:8523 [inline]
    [...] sctp_get_port_local+0x189/0x5a0 net/sctp/socket.c:8270
    [...] sctp_do_bind+0xcc/0x200 net/sctp/socket.c:402
    [...] sctp_bindx_add+0x4b/0xd0 net/sctp/socket.c:497
    [...] sctp_setsockopt_bindx+0x156/0x1b0 net/sctp/socket.c:1022
    [...] sctp_setsockopt net/sctp/socket.c:4641 [inline]
    [...] sctp_setsockopt+0xaea/0x2dc0 net/sctp/socket.c:4611
    [...] sock_common_setsockopt+0x38/0x50 net/core/sock.c:3147
    [...] __sys_setsockopt+0x10f/0x220 net/socket.c:2084
    [...] __do_sys_setsockopt net/socket.c:2100 [inline]

It was caused by when sending msgs without binding a port, in the path:
inet_sendmsg() -> inet_send_prepare() -> inet_autobind() ->
.get_port/sctp_get_port(), sp->bind_hash will be set while bp->port is
not. Later when binding another port by sctp_setsockopt_bindx(), a new
bucket will be created as bp->port is not set.

sctp's autobind is supposed to call sctp_autobind() where it does all
things including setting bp->port. Since sctp_autobind() is called in
sctp_sendmsg() if the sk is not yet bound, it should have skipped the
auto bind.

THis patch is to avoid calling inet_autobind() in inet_send_prepare()
by changing sctp_prot .no_autobind with true, also remove the unused
.get_port.

Reported-by: [email protected]
Signed-off-by: Xin Long <[email protected]>
Acked-by: Marcelo Ricardo Leitner <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

sched: etf: Fix ordering of packets with same txtime

When a application sends many packets with the same txtime, they may
be transmitted out of order (different from the order in which they
were enqueued).

This happens because when inserting elements into the tree, when the
txtime of two packets are the same, the new packet is inserted at the
left side of the tree, causing the reordering. The only effect of this
change should be that packets with the same txtime will be transmitted
in the order they are enqueued.

The application in question (the AVTP GStreamer plugin, still in
development) is sending video traffic, in which each video frame have
a single presentation time, the problem is that when packetizing,
multiple packets end up with the same txtime.

The receiving side was rejecting packets because they were being
received out of order.

Fixes: 25db26a91364 ("net/sched: Introduce the ETF Qdisc")
Reported-by: Ederson de Souza <[email protected]>
Signed-off-by: Vinicius Costa Gomes <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

net: avoid potential infinite loop in tc_ctl_action()

tc_ctl_action() has the ability to loop forever if tcf_action_add()
returns -EAGAIN.

This special case has been done in case a module needed to be loaded,
but it turns out that tcf_add_notify() could also return -EAGAIN
if the socket sk_rcvbuf limit is hit.

We need to separate the two cases, and only loop for the module
loading case.

While we are at it, add a limit of 10 attempts since unbounded
loops are always scary.

syzbot repro was something like :

socket(PF_NETLINK, SOCK_RAW|SOCK_NONBLOCK, NETLINK_ROUTE) = 3
write(3, ..., 38) = 38
setsockopt(3, SOL_SOCKET, SO_RCVBUF, [0], 4) = 0
sendmsg(3, {msg_name(0)=NULL, msg_iov(1)=[{..., 388}], msg_controllen=0, msg_flags=0x10}, ...)

NMI backtrace for cpu 0
CPU: 0 PID: 1054 Comm: khungtaskd Not tainted 5.4.0-rc1+ #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
 __dump_stack lib/dump_stack.c:77 [inline]
 dump_stack+0x172/0x1f0 lib/dump_stack.c:113
 nmi_cpu_backtrace.cold+0x70/0xb2 lib/nmi_backtrace.c:101
 nmi_trigger_cpumask_backtrace+0x23b/0x28b lib/nmi_backtrace.c:62
 arch_trigger_cpumask_backtrace+0x14/0x20 arch/x86/kernel/apic/hw_nmi.c:38
 trigger_all_cpu_backtrace include/linux/nmi.h:146 [inline]
 check_hung_uninterruptible_tasks kernel/hung_task.c:205 [inline]
 watchdog+0x9d0/0xef0 kernel/hung_task.c:289
 kthread+0x361/0x430 kernel/kthread.c:255
 ret_from_fork+0x24/0x30 arch/x86/entry/entry_64.S:352
Sending NMI from CPU 0 to CPUs 1:
NMI backtrace for cpu 1
CPU: 1 PID: 8859 Comm: syz-executor910 Not tainted 5.4.0-rc1+ #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
RIP: 0010:arch_local_save_flags arch/x86/include/asm/paravirt.h:751 [inline]
RIP: 0010:lockdep_hardirqs_off+0x1df/0x2e0 kernel/locking/lockdep.c:3453
Code: 5c 08 00 00 5b 41 5c 41 5d 5d c3 48 c7 c0 58 1d f3 88 48 ba 00 00 00 00 00 fc ff df 48 c1 e8 03 80 3c 10 00 0f 85 d3 00 00 00 <48> 83 3d 21 9e 99 07 00 0f 84 b9 00 00 00 9c 58 0f 1f 44 00 00 f6
RSP: 0018:ffff8880a6f3f1b8 EFLAGS: 00000046
RAX: 1ffffffff11e63ab RBX: ffff88808c9c6080 RCX: 0000000000000000
RDX: dffffc0000000000 RSI: 0000000000000000 RDI: ffff88808c9c6914
RBP: ffff8880a6f3f1d0 R08: ffff88808c9c6080 R09: fffffbfff16be5d1
R10: fffffbfff16be5d0 R11: 0000000000000003 R12: ffffffff8746591f
R13: ffff88808c9c6080 R14: ffffffff8746591f R15: 0000000000000003
FS:  00000000011e4880(0000) GS:ffff8880ae900000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: ffffffffff600400 CR3: 00000000a8920000 CR4: 00000000001406e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
 trace_hardirqs_off+0x62/0x240 kernel/trace/trace_preemptirq.c:45
 __raw_spin_lock_irqsave include/linux/spinlock_api_smp.h:108 [inline]
 _raw_spin_lock_irqsave+0x6f/0xcd kernel/locking/spinlock.c:159
 __wake_up_common_lock+0xc8/0x150 kernel/sched/wait.c:122
 __wake_up+0xe/0x10 kernel/sched/wait.c:142
 netlink_unlock_table net/netlink/af_netlink.c:466 [inline]
 netlink_unlock_table net/netlink/af_netlink.c:463 [inline]
 netlink_broadcast_filtered+0x705/0xb80 net/netlink/af_netlink.c:1514
 netlink_broadcast+0x3a/0x50 net/netlink/af_netlink.c:1534
 rtnetlink_send+0xdd/0x110 net/core/rtnetlink.c:714
 tcf_add_notify net/sched/act_api.c:1343 [inline]
 tcf_action_add+0x243/0x370 net/sched/act_api.c:1362
 tc_ctl_action+0x3b5/0x4bc net/sched/act_api.c:1410
 rtnetlink_rcv_msg+0x463/0xb00 net/core/rtnetlink.c:5386
 netlink_rcv_skb+0x177/0x450 net/netlink/af_netlink.c:2477
 rtnetlink_rcv+0x1d/0x30 net/core/rtnetlink.c:5404
 netlink_unicast_kernel net/netlink/af_netlink.c:1302 [inline]
 netlink_unicast+0x531/0x710 net/netlink/af_netlink.c:1328
 netlink_sendmsg+0x8a5/0xd60 net/netlink/af_netlink.c:1917
 sock_sendmsg_nosec net/socket.c:637 [inline]
 sock_sendmsg+0xd7/0x130 net/socket.c:657
 ___sys_sendmsg+0x803/0x920 net/socket.c:2311
 __sys_sendmsg+0x105/0x1d0 net/socket.c:2356
 __do_sys_sendmsg net/socket.c:2365 [inline]
 __se_sys_sendmsg net/socket.c:2363 [inline]
 __x64_sys_sendmsg+0x78/0xb0 net/socket.c:2363
 do_syscall_64+0xfa/0x760 arch/x86/entry/common.c:290
 entry_SYSCALL_64_after_hwframe+0x49/0xbe
RIP: 0033:0x440939

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Signed-off-by: Eric Dumazet <[email protected]>
Reported-by: [email protected]
Signed-off-by: David S. Miller <[email protected]>

net: dsa: sja1105: Use the correct style for SPDX License Identifier

This patch corrects the SPDX License Identifier style
in header files related to Distributed Switch Architecture
drivers for NXP SJA1105 series Ethernet switch support.
It uses an expilict block comment for the SPDX License
Identifier.

Changes made by using a script provided by Joe Perches here:
https://lkml.org/lkml/2019/2/7/46.

Suggested-by: Joe Perches <[email protected]>
Signed-off-by: Nishad Kamdar <[email protected]>
Reviewed-by: Andrew Lunn <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

tcp: fix a possible lockdep splat in tcp_done()

syzbot found that if __inet_inherit_port() returns an error,
we call tcp_done() after inet_csk_prepare_forced_close(),
meaning the socket lock is no longer held.

We might fix this in a different way in net-next, but
for 5.4 it seems safer to relax the lockdep check.

Fixes: d983ea6f16b8 ("tcp: add rcu protection around tp->fastopen_rsk")
Signed-off-by: Eric Dumazet <[email protected]>
Reported-by: syzbot <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge branch 'Update-MT7629-to-support-PHYLINK-API'

MarkLee says:

====================
Update MT7629 to support PHYLINK API

This patch set has two goals :
1. Fix mt7629 GMII mode issue after apply mediatek
   PHYLINK support patch.
2. Update mt7629 dts to reflect the latest dt-binding
   with PHYLINK support.
====================

Signed-off-by: David S. Miller <[email protected]>

arm: dts: mediatek: Update mt7629 dts to reflect the latest dt-binding

* Removes mediatek,physpeed property from dtsi that is useless in PHYLINK
* Use the fixed-link property speed = <2500> to set the phy in 2.5Gbit.
* Set gmac1 to gmii mode that connect to a internal gphy

Signed-off-by: MarkLee <[email protected]>
Signed-off-by: David S. Miller <[email protected]>