The SQPOLL thread dictates the lock order, and we hold the ctx->uring_lock
for all the registration opcodes. We also hold a ref to the ctx, and we
do drop the lock for other reasons to quiesce, so it's fine to drop the
ctx lock temporarily to grab the sqd->lock. This fixes the following
lockdep splat:

======================================================
WARNING: possible circular locking dependency detected
5.14.0-syzkaller #0 Not tainted
------------------------------------------------------
syz-executor.5/25433 is trying to acquire lock:
ffff888023426870 (&sqd->lock){+.+.}-{3:3}, at: io_register_iowq_max_workers fs/io_uring.c:10551 [inline]
ffff888023426870 (&sqd->lock){+.+.}-{3:3}, at: __io_uring_register fs/io_uring.c:10757 [inline]
ffff888023426870 (&sqd->lock){+.+.}-{3:3}, at: __do_sys_io_uring_register+0x10aa/0x2e70 fs/io_uring.c:10792

but task is already holding lock:
ffff8880885b40a8 (&ctx->uring_lock){+.+.}-{3:3}, at: __do_sys_io_uring_register+0x2e1/0x2e70 fs/io_uring.c:10791

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #1 (&ctx->uring_lock){+.+.}-{3:3}:
       __mutex_lock_common kernel/locking/mutex.c:596 [inline]
       __mutex_lock+0x131/0x12f0 kernel/locking/mutex.c:729
       __io_sq_thread fs/io_uring.c:7291 [inline]
       io_sq_thread+0x65a/0x1370 fs/io_uring.c:7368
       ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:295

-> #0 (&sqd->lock){+.+.}-{3:3}:
       check_prev_add kernel/locking/lockdep.c:3051 [inline]
       check_prevs_add kernel/locking/lockdep.c:3174 [inline]
       validate_chain kernel/locking/lockdep.c:3789 [inline]
       __lock_acquire+0x2a07/0x54a0 kernel/locking/lockdep.c:5015
       lock_acquire kernel/locking/lockdep.c:5625 [inline]
       lock_acquire+0x1ab/0x510 kernel/locking/lockdep.c:5590
       __mutex_lock_common kernel/locking/mutex.c:596 [inline]
       __mutex_lock+0x131/0x12f0 kernel/locking/mutex.c:729
       io_register_iowq_max_workers fs/io_uring.c:10551 [inline]
       __io_uring_register fs/io_uring.c:10757 [inline]
       __do_sys_io_uring_register+0x10aa/0x2e70 fs/io_uring.c:10792
       do_syscall_x64 arch/x86/entry/common.c:50 [inline]
       do_syscall_64+0x35/0xb0 arch/x86/entry/common.c:80
       entry_SYSCALL_64_after_hwframe+0x44/0xae

other info that might help us debug this:

 Possible unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock(&ctx->uring_lock);
                               lock(&sqd->lock);
                               lock(&ctx->uring_lock);
  lock(&sqd->lock);

 *** DEADLOCK ***

Fixes: 2e480058 ("io-wq: provide a way to limit max number of workers")
Reported-by: syzbot+97fa56483f69d677969f@syzkaller.appspotmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

In case of !SQPOLL, io_cqring_ev_posted_iopoll() doesn't provide a
memory barrier required by waitqueue_active(&ctx->poll_wait). There is
a wq_has_sleeper(), which does smb_mb() inside, but it's called only for
SQPOLL.

Fixes: 5fd46178 ("io_uring: be smarter about waking multiple CQ ring waiters")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/2982e53bcea2274006ed435ee2a77197107d8a29.1631130542.git.asml.silence@gmail.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

WARNING: CPU: 0 PID: 10392 at fs/io_uring.c:1151 req_ref_put_and_test
fs/io_uring.c:1151 [inline]
WARNING: CPU: 0 PID: 10392 at fs/io_uring.c:1151 req_ref_put_and_test
fs/io_uring.c:1146 [inline]
WARNING: CPU: 0 PID: 10392 at fs/io_uring.c:1151
io_req_complete_post+0xf5b/0x1190 fs/io_uring.c:1794
Modules linked in:
Call Trace:
 tctx_task_work+0x1e5/0x570 fs/io_uring.c:2158
 task_work_run+0xe0/0x1a0 kernel/task_work.c:164
 tracehook_notify_signal include/linux/tracehook.h:212 [inline]
 handle_signal_work kernel/entry/common.c:146 [inline]
 exit_to_user_mode_loop kernel/entry/common.c:172 [inline]
 exit_to_user_mode_prepare+0x232/0x2a0 kernel/entry/common.c:209
 __syscall_exit_to_user_mode_work kernel/entry/common.c:291 [inline]
 syscall_exit_to_user_mode+0x19/0x60 kernel/entry/common.c:302
 do_syscall_64+0x42/0xb0 arch/x86/entry/common.c:86
 entry_SYSCALL_64_after_hwframe+0x44/0xae

When io_wqe_enqueue() -> io_wqe_create_worker() fails, we can't just
call io_run_cancel() to clean up the request, it's already enqueued via
io_wqe_insert_work() and will be executed either by some other worker
during cancellation (e.g. in io_wq_put_and_exit()).
Reported-by: Hao Sun <sunhao.th@gmail.com>
Fixes: 3146cba9 ("io-wq: make worker creation resilient against signals")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/93b9de0fcf657affab0acfd675d4abcd273ee863.1631092071.git.asml.silence@gmail.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

Pull networking fixes and stragglers from Jakub Kicinski:
 "Networking stragglers and fixes, including changes from netfilter,
  wireless and can.

  Current release - regressions:

   - qrtr: revert check in qrtr_endpoint_post(), fixes audio and wifi

   - ip_gre: validate csum_start only on pull

   - bnxt_en: fix 64-bit doorbell operation on 32-bit kernels

   - ionic: fix double use of queue-lock, fix a sleeping in atomic

   - can: c_can: fix null-ptr-deref on ioctl()

   - cs89x0: disable compile testing on powerpc

  Current release - new code bugs:

   - bridge: mcast: fix vlan port router deadlock, consistently disable
     BH

  Previous releases - regressions:

   - dsa: tag_rtl4_a: fix egress tags, only port 0 was working

   - mptcp: fix possible divide by zero

   - netfilter: nft_ct: protect nft_ct_pcpu_template_refcnt with mutex

   - netfilter: socket: icmp6: fix use-after-scope

   - stmmac: fix MAC not working when system resume back with WoL active

  Previous releases - always broken:

   - ip/ip6_gre: use the same logic as SIT interfaces when computing
     v6LL address

   - seg6: set fc_nlinfo in nh_create_ipv4, nh_create_ipv6

   - mptcp: only send extra TCP acks in eligible socket states

   - dsa: lantiq_gswip: fix maximum frame length

   - stmmac: fix overall budget calculation for rxtx_napi

   - bnxt_en: fix firmware version reporting via devlink

   - renesas: sh_eth: add missing barrier to fix freeing wrong tx
     descriptor

  Stragglers:

   - netfilter: conntrack: switch to siphash

   - netfilter: refuse insertion if chain has grown too large

   - ncsi: add get MAC address command to get Intel i210 MAC address"

* tag 'net-5.15-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (76 commits)
  ieee802154: Remove redundant initialization of variable ret
  net: stmmac: fix MAC not working when system resume back with WoL active
  net: phylink: add suspend/resume support
  net: renesas: sh_eth: Fix freeing wrong tx descriptor
  bonding: 3ad: pass parameter bond_params by reference
  cxgb3: fix oops on module removal
  can: c_can: fix null-ptr-deref on ioctl()
  can: rcar_canfd: add __maybe_unused annotation to silence warning
  net: wwan: iosm: Unify IO accessors used in the driver
  net: wwan: iosm: Replace io.*64_lo_hi() with regular accessors
  net: qcom/emac: Replace strlcpy with strscpy
  ip6_gre: Revert "ip6_gre: add validation for csum_start"
  net: hns3: make hclgevf_cmd_caps_bit_map0 and hclge_cmd_caps_bit_map0 static
  selftests/bpf: Test XDP bonding nest and unwind
  bonding: Fix negative jump label count on nested bonding
  MAINTAINERS: add VM SOCKETS (AF_VSOCK) entry
  stmmac: dwmac-loongson:Fix missing return value
  iwlwifi: fix printk format warnings in uefi.c
  net: create netdev->dev_addr assignment helpers
  bnxt_en: Fix possible unintended driver initiated error recovery
  ...

Pull watchdog updates from Wim Van Sebroeck:

 - add Mediatek MT7986 & MT8195 wdt support

 - add Maxim MAX63xx

 - drop bd70528 support

 - rewrite ixp4xx to watchdog framework

 - constify static struct watchdog_ops for sl28cpld_wdt, mpc8xxx_wdt and
   tqmx86

 - introduce watchdog_dev_suspend/resume

 - several fixes and improvements

* tag 'linux-watchdog-5.15-rc1' of git://www.linux-watchdog.org/linux-watchdog:
  dt-bindings: watchdog: Add compatible for Mediatek MT7986
  watchdog: ixp4xx: Rewrite driver to use core
  watchdog: Start watchdog in watchdog_set_last_hw_keepalive only if appropriate
  watchdog: max63xx_wdt: Add device tree probing
  dt-bindings: watchdog: Add Maxim MAX63xx bindings
  watchdog: mediatek: mt8195: add wdt support
  dt-bindings: reset: mt8195: add toprgu reset-controller header file
  watchdog: tqmx86: Constify static struct watchdog_ops
  watchdog: mpc8xxx_wdt: Constify static struct watchdog_ops
  watchdog: sl28cpld_wdt: Constify static struct watchdog_ops
  watchdog: iTCO_wdt: Fix detection of SMI-off case
  watchdog: bcm2835_wdt: consider system-power-controller property
  watchdog: imx2_wdg: notify wdog core to stop ping worker on suspend
  watchdog: introduce watchdog_dev_suspend/resume
  watchdog: Fix NULL pointer dereference when releasing cdev
  watchdog: only run driver set_pretimeout op if device supports it
  watchdog: bd70528 drop bd70528 support

Pull KVM updates from Paolo Bonzini:
 "ARM:
   - Page ownership tracking between host EL1 and EL2
   - Rely on userspace page tables to create large stage-2 mappings
   - Fix incompatibility between pKVM and kmemleak
   - Fix the PMU reset state, and improve the performance of the virtual
     PMU
   - Move over to the generic KVM entry code
   - Address PSCI reset issues w.r.t. save/restore
   - Preliminary rework for the upcoming pKVM fixed feature
   - A bunch of MM cleanups
   - a vGIC fix for timer spurious interrupts
   - Various cleanups

  s390:
   - enable interpretation of specification exceptions
   - fix a vcpu_idx vs vcpu_id mixup

  x86:
   - fast (lockless) page fault support for the new MMU
   - new MMU now the default
   - increased maximum allowed VCPU count
   - allow inhibit IRQs on KVM_RUN while debugging guests
   - let Hyper-V-enabled guests run with virtualized LAPIC as long as
     they do not enable the Hyper-V "AutoEOI" feature
   - fixes and optimizations for the toggling of AMD AVIC (virtualized
     LAPIC)
   - tuning for the case when two-dimensional paging (EPT/NPT) is
     disabled
   - bugfixes and cleanups, especially with respect to vCPU reset and
     choosing a paging mode based on CR0/CR4/EFER
   - support for 5-level page table on AMD processors

  Generic:
   - MMU notifier invalidation callbacks do not take mmu_lock unless
     necessary
   - improved caching of LRU kvm_memory_slot
   - support for histogram statistics
   - add statistics for halt polling and remote TLB flush requests"

* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm: (210 commits)
  KVM: Drop unused kvm_dirty_gfn_invalid()
  KVM: x86: Update vCPU's hv_clock before back to guest when tsc_offset is adjusted
  KVM: MMU: mark role_regs and role accessors as maybe unused
  KVM: MIPS: Remove a "set but not used" variable
  x86/kvm: Don't enable IRQ when IRQ enabled in kvm_wait
  KVM: stats: Add VM stat for remote tlb flush requests
  KVM: Remove unnecessary export of kvm_{inc,dec}_notifier_count()
  KVM: x86/mmu: Move lpage_disallowed_link further "down" in kvm_mmu_page
  KVM: x86/mmu: Relocate kvm_mmu_page.tdp_mmu_page for better cache locality
  Revert "KVM: x86: mmu: Add guest physical address check in translate_gpa()"
  KVM: x86/mmu: Remove unused field mmio_cached in struct kvm_mmu_page
  kvm: x86: Increase KVM_SOFT_MAX_VCPUS to 710
  kvm: x86: Increase MAX_VCPUS to 1024
  kvm: x86: Set KVM_MAX_VCPU_ID to 4*KVM_MAX_VCPUS
  KVM: VMX: avoid running vmx_handle_exit_irqoff in case of emulation
  KVM: x86/mmu: Don't freak out if pml5_root is NULL on 4-level host
  KVM: s390: index kvm->arch.idle_mask by vcpu_idx
  KVM: s390: Enable specification exception interpretation
  KVM: arm64: Trim guest debug exception handling
  KVM: SVM: Add 5-level page table support for SVM
  ...

Pull dmi fix from Jean Delvare.

Unbreak some existing udev/hwdb modalias matches due to misplaced
product_sku field.

* 'dmi-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jdelvare/staging:
  firmware: dmi: Move product_sku info to the end of the modalias

Pull NTB updates from Jon Mason:
 "Bug fixes and clean-ups for Linux v5.15"

* tag 'ntb-5.15' of git://github.com/jonmason/ntb:
  NTB: switch from 'pci_' to 'dma_' API
  ntb: ntb_pingpong: remove redundant initialization of variables msg_data and spad_data
  NTB: perf: Fix an error code in perf_setup_inbuf()
  NTB: Fix an error code in ntb_msit_probe()
  ntb: intel: remove invalid email address in header comment

Pull remoteproc updates from Bjorn Andersson:

 - move the crash recovery worker to the freezable work queue to avoid
   interaction with other drivers during suspend & resume

 - fix a couple of typos in comments

 - add support for handling the audio DSP on SDM660

 - fix a race between the Qualcomm wireless subsystem driver and the
   associated driver for the RF chip

* tag 'rproc-v5.15' of git://git.kernel.org/pub/scm/linux/kernel/git/andersson/remoteproc:
  remoteproc: q6v5_pas: Add sdm660 ADSP PIL compatible
  dt-bindings: remoteproc: qcom: adsp: Add SDM660 ADSP
  remoteproc: use freezable workqueue for crash notifications
  remoteproc: fix kernel doc for struct rproc_ops
  remoteproc: fix an typo in fw_elf_get_class code comments
  remoteproc: qcom: wcnss: Fix race with iris probe

Pull backlight updates from Lee Jones:
 "Fix-ups:
   - Improve bootloader/kernel device handover

  Bug Fixes:
   - Stabilise backlight in ktd253 driver"

* tag 'backlight-next-5.15' of git://git.kernel.org/pub/scm/linux/kernel/git/lee/backlight:
  backlight: pwm_bl: Improve bootloader/kernel device handover
  backlight: ktd253: Stabilize backlight

Pull MFD updates from Lee Jones:
 "Core Frameworks:
   - Add support for registering devices via MFD cells to Simple MFD (I2C)

  New Drivers:
   - Add support for Renesas Synchronization Management Unit (SMU)

  New Device Support:
   - Add support for N5010 to Intel M10 BMC
   - Add support for Cannon Lake to Intel LPSS ACPI
   - Add support for Samsung SSG{1,2} to ST-Ericsson's U8500 family
   - Add support for TQMx110EB and TQMxE40x to TQ-Systems PLD TQMx86

  New Functionality:
   - Add support for GPIO to Intel LPC ICH
   - Add support for Reset to Texas Instruments TPS65086

  Fix-ups:
   - Trivial, sorting, whitespace, renaming, etc; mt6360-core, db8500-prcmu-regs, tqmx86
   - Device Tree fiddling; syscon, axp20x, qcom,pm8008, ti,tps65086, brcm,cru
   - Use proper APIs for IRQ map resolution; ab8500-core, stmpe, tc3589x, wm8994-irq
   - Pass 'supplied-from' property through axp288_fuel_gauge via swnode
   - Remove unused file entry; MAINTAINERS
   - Make interrupt line optional; tps65086
   - Rename db8500-cpuidle driver symbol; db8500-prcmu
   - Remove support for unused hardware; tqmx86
   - Provide a standard LPC clock frequency for unknown boards; tqmx86
   - Remove unused code; ti_am335x_tscadc
   - Use of_iomap() instead of ioremap(); syscon

  Bug Fixes:
   - Clear GPIO IRQ resource flags when no IRQ is set; tqmx86
   - Fix incorrect/misleading frequencies; db8500-prcmu
   - Mitigate namespace clash with other GPIOBASE users"

* tag 'mfd-next-5.15' of git://git.kernel.org/pub/scm/linux/kernel/git/lee/mfd: (31 commits)
  mfd: lpc_sch: Rename GPIOBASE to prevent build error
  mfd: syscon: Use of_iomap() instead of ioremap()
  dt-bindings: mfd: Add Broadcom CRU
  mfd: ti_am335x_tscadc: Delete superfluous error message
  mfd: tqmx86: Assume 24MHz LPC clock for unknown boards
  mfd: tqmx86: Add support for TQ-Systems DMI IDs
  mfd: tqmx86: Add support for TQMx110EB and TQMxE40x
  mfd: tqmx86: Fix typo in "platform"
  mfd: tqmx86: Remove incorrect TQMx90UC board ID
  mfd: tqmx86: Clear GPIO IRQ resource when no IRQ is set
  mfd: simple-mfd-i2c: Add support for registering devices via MFD cells
  mfd/cpuidle: ux500: Rename driver symbol
  mfd: tps65086: Add cell entry for reset driver
  mfd: tps65086: Make interrupt line optional
  dt-bindings: mfd: Convert tps65086.txt to YAML
  MAINTAINERS: Adjust ARM/NOMADIK/Ux500 ARCHITECTURES to file renaming
  mfd: db8500-prcmu: Handle missing FW variant
  mfd: db8500-prcmu: Rename register header
  mfd: axp20x: Add supplied-from property to axp288_fuel_gauge cell
  mfd: Don't use irq_create_mapping() to resolve a mapping
  ...

Pull gpio updates from Bartosz Golaszewski:
 "We mostly have various improvements and refactoring all over the place
  but also some interesting new features - like the virtio GPIO driver
  that allows guest VMs to use host's GPIOs. We also have a new/old GPIO
  driver for rockchip - this one has been split out of the pinctrl
  driver.

  Summary:

   - new driver: gpio-virtio allowing a guest VM running linux to access
     GPIO lines provided by the host

   - split the GPIO driver out of the rockchip pin control driver

   - add support for a new model to gpio-aspeed-sgpio, refactor the
     driver and use generic device property interfaces, improve property
     sanitization

   - add ACPI support to gpio-tegra186

   - improve the code setting the line names to support multiple GPIO
     banks per device

   - constify a bunch of OF functions in the core GPIO code and make the
     declaration for one of the core OF functions we use consistent
     within its header

   - use software nodes in intel_quark_i2c_gpio

   - add support for the gpio-line-names property in gpio-mt7621

   - use the standard GPIO function for setting the GPIO names in
     gpio-brcmstb

   - fix a bunch of leaks and other bugs in gpio-mpc8xxx

   - use generic pm callbacks in gpio-ml-ioh

   - improve resource management and PM handling in gpio-mlxbf2

   - modernize and improve the gpio-dwapb driver

   - coding style improvements in gpio-rcar

   - documentation fixes and improvements

   - update the MAINTAINERS entry for gpio-zynq

   - minor tweaks in several drivers"

* tag 'gpio-updates-for-v5.15' of git://git.kernel.org/pub/scm/linux/kernel/git/brgl/linux: (35 commits)
  gpio: mpc8xxx: Use 'devm_gpiochip_add_data()' to simplify the code and avoid a leak
  gpio: mpc8xxx: Fix a potential double iounmap call in 'mpc8xxx_probe()'
  gpio: mpc8xxx: Fix a resources leak in the error handling path of 'mpc8xxx_probe()'
  gpio: viperboard: remove platform_set_drvdata() call in probe
  gpio: virtio: Add missing mailings lists in MAINTAINERS entry
  gpio: virtio: Fix sparse warnings
  gpio: remove the obsolete MX35 3DS BOARD MC9S08DZ60 GPIO functions
  gpio: max730x: Use the right include
  gpio: Add virtio-gpio driver
  gpio: mlxbf2: Use DEFINE_RES_MEM_NAMED() helper macro
  gpio: mlxbf2: Use devm_platform_ioremap_resource()
  gpio: mlxbf2: Drop wrong use of ACPI_PTR()
  gpio: mlxbf2: Convert to device PM ops
  gpio: dwapb: Get rid of legacy platform data
  mfd: intel_quark_i2c_gpio: Convert GPIO to use software nodes
  gpio: dwapb: Read GPIO base from gpio-base property
  gpio: dwapb: Unify ACPI enumeration checks in get_irq() and configure_irqs()
  gpiolib: Deduplicate forward declaration in the consumer.h header
  MAINTAINERS: update gpio-zynq.yaml reference
  gpio: tegra186: Add ACPI support
  ...

Pull fuse updates from Miklos Szeredi:

 - Allow mounting an active fuse device. Previously the fuse device
   would always be mounted during initialization, and sharing a fuse
   superblock was only possible through mount or namespace cloning

 - Fix data flushing in syncfs (virtiofs only)

 - Fix data flushing in copy_file_range()

 - Fix a possible deadlock in atomic O_TRUNC

 - Misc fixes and cleanups

* tag 'fuse-update-5.15' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/fuse:
  fuse: remove unused arg in fuse_write_file_get()
  fuse: wait for writepages in syncfs
  fuse: flush extending writes
  fuse: truncate pagecache on atomic_o_trunc
  fuse: allow sharing existing sb
  fuse: move fget() to fuse_get_tree()
  fuse: move option checking into fuse_fill_super()
  fuse: name fs_context consistently
  fuse: fix use after free in fuse_read_interrupt()

Pull kgdb updates from Daniel Thompson:
 "Changes for kgdb/kdb this cycle are dominated by a change from Sumit
  that removes as small (256K) private heap from kdb. This is change
  I've hoped for ever since I discovered how few users of this heap
  remained in the kernel, so many thanks to Sumit for hunting these
  down.

  The other change is an incremental step towards SPDX headers"

* tag 'kgdb-5.15-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/danielt/linux:
  kernel: debug: Convert to SPDX identifier
  kdb: Rename members of struct kdbtab_t
  kdb: Simplify kdb_defcmd macro logic
  kdb: Get rid of redundant kdb_register_flags()
  kdb: Rename struct defcmd_set to struct kdb_macro
  kdb: Get rid of custom debug heap allocator

This reverts commit b6558434.

Just like with the memcg lock accounting, the kernel test robot reports
a sizeable performance regression for this commit, and while it clearly
does the rigth thing in theory, we'll need to look at just how to avoid
or minimize the performance overhead of the memcg accounting.

People already have suggestions on how to do that, but it's "future
work".

So revert it for now.

[ Note: the first link below is for this same commit but a different
  commit ID, because it's the kernel test robot ended up noticing it in
  Andrew Morton's patch queue ]

Link: https://lore.kernel.org/lkml/20210905132732.GC15026@xsang-OptiPlex-9020/
Link: https://lore.kernel.org/lkml/20210907150757.GE17617@xsang-OptiPlex-9020/Acked-by: Jens Axboe <axboe@kernel.dk>
Acked-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Roman Gushchin <guro@fb.com>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

This reverts commit 0f12156d.

The kernel test robot reports a sizeable performance regression for this
commit, and while it clearly does the rigth thing in theory, we'll need
to look at just how to avoid or minimize the performance overhead of the
memcg accounting.

People already have suggestions on how to do that, but it's "future
work".

So revert it for now.

Link: https://lore.kernel.org/lkml/20210907150757.GE17617@xsang-OptiPlex-9020/Acked-by: Jens Axboe <axboe@kernel.dk>
Acked-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Roman Gushchin <guro@fb.com>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

This reverts commit 9857a17f.

That commit was completely broken, and I should have caught on to it
earlier.  But happily, the kernel test robot noticed the breakage fairly
quickly.

The breakage is because "try_get_page()" is about avoiding the page
reference count overflow case, but is otherwise the exact same as a
plain "get_page()".

In contrast, "try_get_compound_head()" is an entirely different beast,
and uses __page_cache_add_speculative() because it's not just about the
page reference count, but also about possibly racing with the underlying
page going away.

So all the commentary about how

 "try_get_page() has fallen a little behind in terms of maintenance,
  try_get_compound_head() handles speculative page references more
  thoroughly"

was just completely wrong: yes, try_get_compound_head() handles
speculative page references, but the point is that try_get_page() does
not, and must not.

So there's no lack of maintainance - there are fundamentally different
semantics.

A speculative page reference would be entirely wrong in "get_page()",
and it's entirely wrong in "try_get_page()".  It's not about
speculation, it's purely about "uhhuh, you can't get this page because
you've tried to increment the reference count too much already".

The reason the kernel test robot noticed this bug was that it hit the
VM_BUG_ON() in __page_cache_add_speculative(), which is all about
verifying that the context of any speculative page access is correct.
But since that isn't what try_get_page() is all about, the VM_BUG_ON()
tests things that are not correct to test for try_get_page().
Reported-by: kernel test robot <oliver.sang@intel.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

The variable ret is being initialized with a value that is never read, it
is being updated later on. The assignment is redundant and can be removed.

Addresses-Coverity: ("Unused value")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Joakim Zhang says:

====================
net: stmmac: fix WoL issue

This patch set fixes stmmac not working after system resume back with WoL
active. Thanks a lot for Russell King keeps looking into this issue.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

We can reproduce this issue with below steps:
1) enable WoL on the host
2) host system suspended
3) remote client send out wakeup packets
We can see that host system resume back, but can't work, such as ping failed.

After a bit digging, this issue is introduced by the commit 46f69ded
("net: stmmac: Use resolved link config in mac_link_up()"), which use
the finalised link parameters in mac_link_up() rather than the
parameters in mac_config().

There are two scenarios for MAC suspend/resume in STMMAC driver:

1) MAC suspend with WoL inactive, stmmac_suspend() call
phylink_mac_change() to notify phylink machine that a change in MAC
state, then .mac_link_down callback would be invoked. Further, it will
call phylink_stop() to stop the phylink instance. When MAC resume back,
firstly phylink_start() is called to start the phylink instance, then
call phylink_mac_change() which will finally trigger phylink machine to
invoke .mac_config and .mac_link_up callback. All is fine since
configuration in these two callbacks will be initialized, that means MAC
can restore the state.

2) MAC suspend with WoL active, phylink_mac_change() will put link
down, but there is no phylink_stop() to stop the phylink instance, so it
will link up again, that means .mac_config and .mac_link_up would be
invoked before system suspended. After system resume back, it will do
DMA initialization and SW reset which let MAC lost the hardware setting
(i.e MAC_Configuration register(offset 0x0) is reset). Since link is up
before system suspended, so .mac_link_up would not be invoked after
system resume back, lead to there is no chance to initialize the
configuration in .mac_link_up callback, as a result, MAC can't work any
longer.

After discussed with Russell King [1], we confirm that phylink framework
have not take WoL into consideration yet. This patch calls
phylink_suspend()/phylink_resume() functions which is newly introduced
by Russell King to fix this issue.

[1] https://lore.kernel.org/netdev/20210901090228.11308-1-qiangqing.zhang@nxp.com/

Fixes: 46f69ded ("net: stmmac: Use resolved link config in mac_link_up()")
Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Joakim Zhang reports that Wake-on-Lan with the stmmac ethernet driver broke
when moving the incorrect handling of mac link state out of mac_config().
This reason this breaks is because the stmmac's WoL is handled by the MAC
rather than the PHY, and phylink doesn't cater for that scenario.

This patch adds the necessary phylink code to handle suspend/resume events
according to whether the MAC still needs a valid link or not. This is the
barest minimum for this support.
Reported-by: Joakim Zhang <qiangqing.zhang@nxp.com>
Tested-by: Joakim Zhang <qiangqing.zhang@nxp.com>
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The cur_tx counter must be incremented after TACT bit of
txdesc->status was set. However, a CPU is possible to reorder
instructions and/or memory accesses between cur_tx and
txdesc->status. And then, if TX interrupt happened at such a
timing, the sh_eth_tx_free() may free the descriptor wrongly.
So, add wmb() before cur_tx++.
Otherwise NETDEV WATCHDOG timeout is possible to happen.

Fixes: 86a74ff2 ("net: sh_eth: add support for Renesas SuperH Ethernet")
Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The parameter bond_params is a relatively large 192 byte sized
struct so pass it by reference rather than by value to reduce
copying.

Addresses-Coverity: ("Big parameter passed by value")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge tag 'linux-can-fixes-for-5.15-20210907' of git://git.kernel.org/pub/scm/linux/kernel/git/mkl/linux-can

linux-can-fixes-for-5.15-20210907
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge tag 'wireless-drivers-2021-09-07' of git://git.kernel.org/pub/scm/linux/kernel/git/kvalo/wireless-drivers

Kalle Valo says:

====================
wireless-drivers fixes for v5.15

First set of fixes for v5.15 and only iwlwifi patches this time. Most
important being support for new hardware and new firmware API.

I had already earlier applied a fix which also Linus applied to this
tree as commit 1476ff21 ("iwl: fix debug printf format strings"),
but this doesn't seem to cause any conflicts so I left it there.

iwlwifi

* add support for firmware API 66

* add support for Samsung Galaxy Book Flex2 Alpha

* fix a leak happening every time module is loaded

* fix a printk compiler warning
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

When removing the driver module w/o bringing an interface up before
the error below occurs. Reason seems to be that cancel_work_sync() is
called in t3_sge_stop() for a queue that hasn't been initialized yet.

[10085.941785] ------------[ cut here ]------------
[10085.941799] WARNING: CPU: 1 PID: 5850 at kernel/workqueue.c:3074 __flush_work+0x3ff/0x480
[10085.941819] Modules linked in: vfat snd_hda_codec_hdmi fat snd_hda_codec_realtek snd_hda_codec_generic ledtrig_audio led_class ee1004 iTCO_
wdt intel_tcc_cooling x86_pkg_temp_thermal coretemp aesni_intel crypto_simd cryptd snd_hda_intel snd_intel_dspcfg snd_hda_codec snd_hda_core r
8169 snd_pcm realtek mdio_devres snd_timer snd i2c_i801 i2c_smbus libphy i915 i2c_algo_bit cxgb3(-) intel_gtt ttm mdio drm_kms_helper mei_me s
yscopyarea sysfillrect sysimgblt mei fb_sys_fops acpi_pad sch_fq_codel crypto_user drm efivarfs ext4 mbcache jbd2 crc32c_intel
[10085.941944] CPU: 1 PID: 5850 Comm: rmmod Not tainted 5.14.0-rc7-next-20210826+ #6
[10085.941974] Hardware name: System manufacturer System Product Name/PRIME H310I-PLUS, BIOS 2603 10/21/2019
[10085.941992] RIP: 0010:__flush_work+0x3ff/0x480
[10085.942003] Code: c0 74 6b 65 ff 0d d1 bd 78 75 e8 bc 2f 06 00 48 c7 c6 68 b1 88 8a 48 c7 c7 e0 5f b4 8b 45 31 ff e8 e6 66 04 00 e9 4b fe ff ff <0f> 0b 45 31 ff e9 41 fe ff ff e8 72 c1 79 00 85 c0 74 87 80 3d 22
[10085.942036] RSP: 0018:ffffa1744383fc08 EFLAGS: 00010246
[10085.942048] RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000923
[10085.942062] RDX: 0000000000000000 RSI: 0000000000000001 RDI: ffff91c901710a88
[10085.942076] RBP: ffffa1744383fce8 R08: 0000000000000001 R09: 0000000000000001
[10085.942090] R10: 00000000000000c2 R11: 0000000000000000 R12: ffff91c901710a88
[10085.942104] R13: 0000000000000000 R14: ffff91c909a96100 R15: 0000000000000001
[10085.942118] FS:  00007fe417837740(0000) GS:ffff91c969d00000(0000) knlGS:0000000000000000
[10085.942134] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[10085.942146] CR2: 000055a8d567ecd8 CR3: 0000000121690003 CR4: 00000000003706e0
[10085.942160] Call Trace:
[10085.942166]  ? __lock_acquire+0x3af/0x22e0
[10085.942177]  ? cancel_work_sync+0xb/0x10
[10085.942187]  __cancel_work_timer+0x128/0x1b0
[10085.942197]  ? __pm_runtime_resume+0x5b/0x90
[10085.942208]  cancel_work_sync+0xb/0x10
[10085.942217]  t3_sge_stop+0x2f/0x50 [cxgb3]
[10085.942234]  remove_one+0x26/0x190 [cxgb3]
[10085.942248]  pci_device_remove+0x39/0xa0
[10085.942258]  __device_release_driver+0x15e/0x240
[10085.942269]  driver_detach+0xd9/0x120
[10085.942278]  bus_remove_driver+0x53/0xd0
[10085.942288]  driver_unregister+0x2c/0x50
[10085.942298]  pci_unregister_driver+0x31/0x90
[10085.942307]  cxgb3_cleanup_module+0x10/0x18c [cxgb3]
[10085.942324]  __do_sys_delete_module+0x191/0x250
[10085.942336]  ? syscall_enter_from_user_mode+0x21/0x60
[10085.942347]  ? trace_hardirqs_on+0x2a/0xe0
[10085.942357]  __x64_sys_delete_module+0x13/0x20
[10085.942368]  do_syscall_64+0x40/0x90
[10085.942377]  entry_SYSCALL_64_after_hwframe+0x44/0xae
[10085.942389] RIP: 0033:0x7fe41796323b

Fixes: 5e0b8928 ("net:cxgb3: replace tasklets with works")
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

One MIPS platform (mach-rc32434) defines GPIOBASE. This macro
conflicts with one of the same name in lpc_sch.c. Rename the latter one
to prevent the build error.

../drivers/mfd/lpc_sch.c:25: error: "GPIOBASE" redefined [-Werror]
   25 | #define GPIOBASE        0x44
../arch/mips/include/asm/mach-rc32434/rb.h:32: note: this is the location of the previous definition
   32 | #define GPIOBASE        0x050000

Cc: Denis Turischev <denis@compulab.co.il>
Fixes: e82c60ae ("mfd: Introduce lpc_sch for Intel SCH LPC bridge")
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Lee Jones <lee.jones@linaro.org>

This automatically selects between ioremap() and ioremap_np() on
platforms that require it, such as Apple SoCs.
Signed-off-by: Hector Martin <marcan@marcan.st>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Lee Jones <lee.jones@linaro.org>

The pdev maybe not a platform device, e.g. c_can_pci device, in this
case, calling to_platform_device() would not make sense. Also, per the
comment in drivers/net/can/c_can/c_can_ethtool.c, @bus_info should
match dev_name() string, so I am replacing this with dev_name() to fix
this issue.

[    1.458583] BUG: unable to handle page fault for address: 0000000100000000
[    1.460921] RIP: 0010:strnlen+0x1a/0x30
[    1.466336]  ? c_can_get_drvinfo+0x65/0xb0 [c_can]
[    1.466597]  ethtool_get_drvinfo+0xae/0x360
[    1.466826]  dev_ethtool+0x10f8/0x2970
[    1.467880]  sock_ioctl+0xef/0x300

Fixes: 2722ac98 ("can: c_can: add ethtool support")
Link: https://lore.kernel.org/r/20210906233704.1162666-1-ztong0001@gmail.com
Cc: stable@vger.kernel.org # 5.14+
Signed-off-by: Tong Zhang <ztong0001@gmail.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

Since commit

| dd3bd23e ("can: rcar_canfd: Add Renesas R-Car CAN FD driver")

the rcar_canfd driver can be compile tested on all architectures. On
non OF enabled archs, or archs where OF is optional (and disabled in
the .config) the compilation throws the following warning:

| drivers/net/can/rcar/rcar_canfd.c:2020:34: warning: unused variable 'rcar_canfd_of_table' [-Wunused-const-variable]
| static const struct of_device_id rcar_canfd_of_table[] = {
|                                  ^

This patch fixes the warning by marking the variable
rcar_canfd_of_table as __maybe_unused.

Fixes: ac422408 ("can: rcar: Kconfig: Add helper dependency on COMPILE_TEST")
Fixes: dd3bd23e ("can: rcar_canfd: Add Renesas R-Car CAN FD driver")
Link: https://lore.kernel.org/all/20210907064537.1054268-1-mkl@pengutronix.de
Cc: linux-renesas-soc@vger.kernel.org
Cc: Cai Huoqing <caihuoqing@baidu.com>
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

The tb_test_credit_alloc_all() function had a huge number of
KUNIT_ASSERT() statements, all of which (though the magic of many many
layers of inscrutable macros) ended up allocating and initializing
various test assertion structures on the stack.

Don't do that.  The kernel stack isn't infinite, and we have compiler
warnings (now errors) for the case where a stack frame grows too large.

Like it did here, by not an inconsiderable margin:

   drivers/thunderbolt/test.c: In function ‘tb_test_credit_alloc_all’:
   drivers/thunderbolt/test.c:2367:1: error: the frame size of 4500 bytes is larger than 2048 bytes [-Werror=frame-larger-than=]
    2367 | }
         | ^

Solve this similarly to the lib/test_scanf case: split out the tests
into several smaller functions, each just testing one particular tunnel
credit allocation.

This makes the i386 allyesconfig build work for me again.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

It turns out that gcc has real trouble merging all the temporary
on-stack buffer allocation.  So despite the fact that their lifetimes do
not overlap, gcc will allocate stack for all of them when they have
different types.  Which they do in the number scanning test routines.

This is unfortunate in general, but with lots of test-cases in one
function, it becomes a real problem.  gcc will allocate a huge stack
frame for no actual good reason.

We have tried to counteract this tendency of gcc not merging stack slots
(see "-fconserve-stack"), but that has limited effect (and should be on
by default these days, iirc).

So with all the debug options enabled on an i386 allmodconfig build, we
end up with overly big stack frames, and the resulting stack frame size
warnings (now errors):

   lib/test_scanf.c: In function ‘numbers_list_field_width_val_width’:
   lib/test_scanf.c:530:1: error: the frame size of 2088 bytes is larger than 2048 bytes [-Werror=frame-larger-than=]
     530 | }
         | ^
   lib/test_scanf.c: In function ‘numbers_list_field_width_typemax’:
   lib/test_scanf.c:488:1: error: the frame size of 2568 bytes is larger than 2048 bytes [-Werror=frame-larger-than=]
     488 | }
         | ^
   lib/test_scanf.c: In function ‘numbers_list’:
   lib/test_scanf.c:437:1: error: the frame size of 2088 bytes is larger than 2048 bytes [-Werror=frame-larger-than=]
     437 | }
         | ^

In this particular case, the reasonably straightforward solution is to
just split out the test routines into multiple more targeted versions.
That way we don't have one huge stack, but several smaller ones, and
they aren't active all at the same time.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

The variable 'package_size' is an unsigned long, and should be printed
out using '%lu', not '%zd' (that would be for a size_t).

Yes, on many architectures (including x86-64), 'size_t' is in fact the
same type as 'long', but that's a fairly random architecture definition,
and on some platforms 'size_t' is in fact 'int' rather than 'long'.

That is the case on traditional 32-bit x86.  Yes, both types are the
exact same 32-bit size, and it would all print out perfectly correctly,
but '%zd' ends up still being wrong.

And we can't make 'package_size' be a 'size_t', because we get the
actual value using efivar_entry_get() that takes a pointer to an
'unsigned long'.  So '%lu' it is.

This fixes two of the i386 allmodconfig build warnings (that is now an
error due to -Werror).
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Pull block fixes from Jens Axboe:
 "Was going to send this one in later this week, but given that -Werror
  is now enabled (or at least available), the mq-deadline fix really
  should go in for the folks hitting that.

   - Ensure dd_queued() is only there if needed (Geert)

   - Fix a kerneldoc warning for bio_alloc_kiocb()

   - BFQ fix for queue merging

   - loop locking fix (Tetsuo)"

* tag 'block-5.15-2021-09-05' of git://git.kernel.dk/linux-block:
  loop: reduce the loop_ctl_mutex scope
  bio: fix kerneldoc documentation for bio_alloc_kiocb()
  block, bfq: honor already-setup queue merges
  block/mq-deadline: Move dd_queued() to fix defined but not used warning

Pull CDROM maintainer update from Jens Axboe:
 "It's been about 22 years since I originally started maintaining the
  CDROM code, and I just haven't been able to even get reviews done in a
  timely fashion the last handful of years.

  Time to pass it on, and Phillip has volunteered take over these
  duties. I'll be helping as needed for the foreseeable future"

* tag 'misc-5.15-2021-09-05' of git://git.kernel.dk/linux-block:
  cdrom: update uniform CD-ROM maintainership in MAINTAINERS file

Pull libata fixes from Jens Axboe:
 "Fixes for queued trim on certain Samsung SSDs, in conjunction with
  certain ATI controllers"

* tag 'libata-5.15-2021-09-05' of git://git.kernel.dk/linux-block:
  libata: Add ATA_HORKAGE_NO_NCQ_ON_ATI for Samsung 860 and 870 SSD.
  libata: add ATA_HORKAGE_NO_NCQ_TRIM for Samsung 860 and 870 SSDs

Pull io_uring fixes from Jens Axboe:
 "As sometimes happens, two reports came in around the merge window open
  that led to some fixes. Hence this one is a bit bigger than usual
  followup fixes, but most of it will be going towards stable, outside
  of the fixes that are addressing regressions from this merge window.

  In detail:

   - postgres is a heavy user of signals between tasks, and if we're
     unlucky this can interfere with io-wq worker creation. Make sure
     we're resilient against unrelated signal handling. This set of
     changes also includes hardening against allocation failures, which
     could previously had led to stalls.

   - Some use cases that end up having a mix of bounded and unbounded
     work would have starvation issues related to that. Split the
     pending work lists to handle that better.

   - Completion trace int -> unsigned -> long fix

   - Fix issue with REGISTER_IOWQ_MAX_WORKERS and SQPOLL

   - Fix regression with hash wait lock in this merge window

   - Fix retry issued on block devices (Ming)

   - Fix regression with links in this merge window (Pavel)

   - Fix race with multi-shot poll and completions (Xiaoguang)

   - Ensure regular file IO doesn't inadvertently skip completion
     batching (Pavel)

   - Ensure submissions are flushed after running task_work (Pavel)"

* tag 'for-5.15/io_uring-2021-09-04' of git://git.kernel.dk/linux-block:
  io_uring: io_uring_complete() trace should take an integer
  io_uring: fix possible poll event lost in multi shot mode
  io_uring: prolong tctx_task_work() with flushing
  io_uring: don't disable kiocb_done() CQE batching
  io_uring: ensure IORING_REGISTER_IOWQ_MAX_WORKERS works with SQPOLL
  io-wq: make worker creation resilient against signals
  io-wq: get rid of FIXED worker flag
  io-wq: only exit on fatal signals
  io-wq: split bounded and unbounded work into separate lists
  io-wq: fix queue stalling race
  io_uring: don't submit half-prepared drain request
  io_uring: fix queueing half-created requests
  io-wq: ensure that hash wait lock is IRQ disabling
  io_uring: retry in case of short read on block device
  io_uring: IORING_OP_WRITE needs hash_reg_file set
  io-wq: fix race between adding work and activating a free worker

Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Currently we have readl()/writel()/ioread*()/iowrite*() APIs in use.
Let's unify to use only ioread*()/iowrite*() variants.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The io.*_lo_hi() variants are not strictly needed on the x86 hardware
and especially the PCI bus. Replace them with regular accessors, but
leave headers in place in case of 32-bit build.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>