If iopl() is disabled, then providing ioperm() does not make much sense.

Rename the config option and disable/enable both syscalls with it. Guard
the code with #ifdefs where appropriate.
Suggested-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

The IOPL emulation via the I/O bitmap is sufficient. Remove the legacy
cruft dealing with the (e)flags based IOPL mechanism.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Juergen Gross <jgross@suse.com> (Paravirt and Xen parts)
Acked-by: Andy Lutomirski <luto@kernel.org>

The access to the full I/O port range can be also provided by the TSS I/O
bitmap, but that would require to copy 8k of data on scheduling in the
task. As shown with the sched out optimization TSS.io_bitmap_base can be
used to switch the incoming task to a preallocated I/O bitmap which has all
bits zero, i.e. allows access to all I/O ports.

Implementing this allows to provide an iopl() emulation mode which restricts
the IOPL level 3 permissions to I/O port access but removes the STI/CLI
permission which is coming with the hardware IOPL mechansim.

Provide a config option to switch IOPL to emulation mode, make it the
default and while at it also provide an option to disable IOPL completely.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Andy Lutomirski <luto@kernel.org>

The comment for the sys_iopl() implementation is outdated and actively
misleading in some parts. Fix it up.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Andy Lutomirski <luto@kernel.org>

Add code to the fork path which forces the shared bitmap to be duplicated
and the reference count to be dropped. Verify that the child modifications
did not affect the parent.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

The I/O bitmap is duplicated on fork. That's wasting memory and slows down
fork. There is no point to do so. As long as the bitmap is not modified it
can be shared between threads and processes.

Add a refcount and just share it on fork. If a task modifies the bitmap
then it has to do the duplication if and only if it is shared.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Andy Lutomirski <luto@kernel.org>

If ioperm() results in a bitmap with all bits set (no permissions to any
I/O port), then handling that bitmap on context switch and exit to user
mode is pointless. Drop it.

Move the bitmap exit handling to the ioport code and reuse it for both the
thread exit path and dropping it. This allows to reuse this code for the
upcoming iopl() emulation.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Andy Lutomirski <luto@kernel.org>

There is no point to update the TSS bitmap for tasks which use I/O bitmaps
on every context switch. It's enough to update it right before exiting to
user space.

That reduces the context switch bitmap handling to invalidating the io
bitmap base offset in the TSS when the outgoing task has TIF_IO_BITMAP
set. The invaldiation is done on purpose when a task with an IO bitmap
switches out to prevent any possible leakage of an activated IO bitmap.

It also removes the requirement to update the tasks bitmap atomically in
ioperm().
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

Add a globally unique sequence number which is incremented when ioperm() is
changing the I/O bitmap of a task. Store the new sequence number in the
io_bitmap structure and compare it with the sequence number of the I/O
bitmap which was last loaded on a CPU. Only update the bitmap if the
sequence is different.

That should further reduce the overhead of I/O bitmap scheduling when there
are only a few I/O bitmap users on the system.

The 64bit sequence counter is sufficient. A wraparound of the sequence
counter assuming an ioperm() call every nanosecond would require about 584
years of uptime.
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

No point in having all the data in thread_struct, especially as upcoming
changes add more.

Make the bitmap in the new struct accessible as array of longs and as array
of characters via a union, so both the bitmap functions and the update
logic can avoid type casts.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

Move the non hardware portion of I/O bitmap data into a seperate struct for
readability sake.
Originally-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

There is no requirement to update the TSS I/O bitmap when a thread using it is
scheduled out and the incoming thread does not use it.

For the permission check based on the TSS I/O bitmap the CPU calculates the memory
location of the I/O bitmap by the address of the TSS and the io_bitmap_base member
of the tss_struct. The easiest way to invalidate the I/O bitmap is to switch the
offset to an address outside of the TSS limit.

If an I/O instruction is issued from user space the TSS limit causes #GP to be
raised in the same was as valid I/O bitmap with all bits set to 1 would do.

This removes the extra work when an I/O bitmap using task is scheduled out
and puts the burden on the rare I/O bitmap users when they are scheduled
in.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

If ioperm() is invoked the first time and the @turn_on argument is 0, then
there is no point to allocate a bitmap just to clear permissions which are
not set.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

On the first allocation of a task the I/O bitmap needs to be
allocated. After the allocation it is installed as an empty bitmap and
immediately afterwards updated.

Avoid that and just do the initial updates (store bitmap pointer, set TIF
flag and make TSS limit valid) in the update path unconditionally. If the
bitmap was already active this is redundant but harmless.

Preparatory change for later optimizations in the context switch code.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

Get rid of superfluous includes.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Andy Lutomirski <luto@kernel.org>

The BUILD_BUG_ON(IO_BITMAP_OFFSET - 1 == 0x67) in the VMX code is bogus in
two aspects:

1) This wants to be in generic x86 code simply to catch issues even when
   VMX is disabled in Kconfig.

2) The IO_BITMAP_OFFSET is not the right thing to check because it makes
   asssumptions about the layout of tss_struct. Nothing requires that the
   I/O bitmap is placed right after x86_tss, which is the hardware mandated
   tss structure. It pointlessly makes restrictions on the struct
   tss_struct layout.

The proper thing to check is:

    - Offset of x86_tss in tss_struct is 0
    - Size of x86_tss == 0x68

Move it to the other build time TSS checks and make it do the right thing.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Acked-by: Andy Lutomirski <luto@kernel.org>

Similar to copy_thread_tls() the 32bit and 64bit implementations of
cpu_init() are very similar and unification avoids duplicate changes in the
future.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Andy Lutomirski <luto@kernel.org>

While looking at the TSS io bitmap it turned out that any change in that
area would require identical changes to copy_thread_tls(). The 32 and 64
bit variants share sufficient code to consolidate them into a common
function to avoid duplication of upcoming modifications.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Andy Lutomirski <luto@kernel.org>

The active() callback of the IO bitmap regset divides the IO bitmap size by
the word size (32/64 bit). As the I/O bitmap size is in bytes the active
check fails for bitmap sizes of 1-3 bytes on 32bit and 1-7 bytes on 64bit.

Use DIV_ROUND_UP() instead.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Andy Lutomirski <luto@kernel.org>

Pull ARM SoC fixes from Olof Johansson:
 "A set of fixes that have trickled in over the last couple of weeks:

   - MAINTAINER update for Cavium/Marvell ThunderX2

   - stm32 tweaks to pinmux for Joystick/Camera, and RAM allocation for
     CAN interfaces

   - i.MX fixes for voltage regulator GPIO mappings, fixes voltage
     scaling issues

   - More i.MX fixes for various issues on i.MX eval boards: interrupt
     storm due to u-boot leaving pins in new states, fixing power button
     config, a couple of compatible-string corrections.

   - Powerdown and Suspend/Resume fixes for Allwinner A83-based tablets

   - A few documentation tweaks and a fix of a memory leak in the reset
     subsystem"

* tag 'armsoc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc:
  MAINTAINERS: update Cavium ThunderX2 maintainers
  ARM: dts: stm32: change joystick pinctrl definition on stm32mp157c-ev1
  ARM: dts: stm32: remove OV5640 pinctrl definition on stm32mp157c-ev1
  ARM: dts: stm32: Fix CAN RAM mapping on stm32mp157c
  ARM: dts: stm32: relax qspi pins slew-rate for stm32mp157
  arm64: dts: zii-ultra: fix ARM regulator GPIO handle
  ARM: sunxi: Fix CPU powerdown on A83T
  ARM: dts: sun8i-a83t-tbs-a711: Fix WiFi resume from suspend
  arm64: dts: imx8mn: fix compatible string for sdma
  arm64: dts: imx8mm: fix compatible string for sdma
  reset: fix reset_control_ops kerneldoc comment
  ARM: dts: imx6-logicpd: Re-enable SNVS power key
  soc: imx: gpc: fix initialiser format
  ARM: dts: imx6qdl-sabreauto: Fix storm of accelerometer interrupts
  arm64: dts: ls1028a: fix a compatible issue
  reset: fix reset_control_get_exclusive kerneldoc comment
  reset: fix reset_control_lookup kerneldoc comment
  reset: fix of_reset_control_get_count kerneldoc comment
  reset: fix of_reset_simple_xlate kerneldoc comment
  reset: Fix memory leak in reset_control_array_put()

Pull IIO fixes and staging driver from Greg KH:
 "Here is a mix of a number of IIO driver fixes for 5.4-rc7, and a whole
  new staging driver.

  The IIO fixes resolve some reported issues, all are tiny.

  The staging driver addition is the vboxsf filesystem, which is the
  VirtualBox guest shared folder code. Hans has been trying to get
  filesystem reviewers to review the code for many months now, and
  Christoph finally said to just merge it in staging now as it is
  stand-alone and the filesystem people can review it easier over time
  that way.

  I know it's late for this big of an addition, but it is stand-alone.

  The code has been in linux-next for a while, long enough to pick up a
  few tiny fixes for it already so people are looking at it.

  All of these have been in linux-next with no reported issues"

* tag 'staging-5.4-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging:
  staging: Fix error return code in vboxsf_fill_super()
  staging: vboxsf: fix dereference of pointer dentry before it is null checked
  staging: vboxsf: Remove unused including <linux/version.h>
  staging: Add VirtualBox guest shared folder (vboxsf) support
  iio: adc: stm32-adc: fix stopping dma
  iio: imu: inv_mpu6050: fix no data on MPU6050
  iio: srf04: fix wrong limitation in distance measuring
  iio: imu: adis16480: make sure provided frequency is positive

Pull char/misc driver fixes from Greg KH:
 "Here are a number of late-arrival driver fixes for issues reported for
  some char/misc drivers for 5.4-rc7

  These all come from the different subsystem/driver maintainers as
  things that they had reports for and wanted to see fixed.

  All of these have been in linux-next with no reported issues"

* tag 'char-misc-5.4-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/char-misc:
  intel_th: pci: Add Jasper Lake PCH support
  intel_th: pci: Add Comet Lake PCH support
  intel_th: msu: Fix possible memory leak in mode_store()
  intel_th: msu: Fix overflow in shift of an unsigned int
  intel_th: msu: Fix missing allocation failure check on a kstrndup
  intel_th: msu: Fix an uninitialized mutex
  intel_th: gth: Fix the window switching sequence
  soundwire: slave: fix scanf format
  soundwire: intel: fix intel_register_dai PDI offsets and numbers
  interconnect: Add locking in icc_set_tag()
  interconnect: qcom: Fix icc_onecell_data allocation
  soundwire: depend on ACPI || OF
  soundwire: depend on ACPI
  thunderbolt: Drop unnecessary read when writing LC command in Ice Lake
  thunderbolt: Fix lockdep circular locking depedency warning
  thunderbolt: Read DP IN adapter first two dwords in one go

Pull configfs regression fix from Christoph Hellwig:
 "Fix a regression from this merge window in the configfs symlink
  handling (Honggang Li)"

* tag 'configfs-for-5.4-2' of git://git.infradead.org/users/hch/configfs:
  configfs: calculate the depth of parent item

Pull x86 fixes from Thomas Gleixner:
 "A small set of fixes for x86:

   - Make the tsc=reliable/nowatchdog command line parameter work again.
     It was broken with the introduction of the early TSC clocksource.

   - Prevent the evaluation of exception stacks before they are set up.
     This causes a crash in dumpstack because the stack walk termination
     gets screwed up.

   - Prevent a NULL pointer dereference in the rescource control file
     system.

   - Avoid bogus warnings about APIC id mismatch related to the LDR
     which can happen when the LDR is not in use and therefore not
     initialized. Only evaluate that when the APIC is in logical
     destination mode"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/tsc: Respect tsc command line paraemeter for clocksource_tsc_early
  x86/dumpstack/64: Don't evaluate exception stacks before setup
  x86/apic/32: Avoid bogus LDR warnings
  x86/resctrl: Prevent NULL pointer dereference when reading mondata

Pull timer fixes from Thomas Gleixner:
 "A small set of fixes for timekeepoing and clocksource drivers:

   - VDSO data was updated conditional on the availability of a VDSO
     capable clocksource. This causes the VDSO functions which do not
     depend on a VDSO capable clocksource to operate on stale data.
     Always update unconditionally.

   - Prevent a double free in the mediatek driver

   - Use the proper helper in the sh_mtu2 driver so it won't attempt to
     initialize non-existing interrupts"

* 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  timekeeping/vsyscall: Update VDSO data unconditionally
  clocksource/drivers/sh_mtu2: Do not loop using platform_get_irq_by_name()
  clocksource/drivers/mediatek: Fix error handling

Pull scheduler fixes from Thomas Gleixner:
 "Two fixes for scheduler regressions:

   - Plug a subtle race condition which was introduced with the rework
     of the next task selection functionality. The change of task
     properties became unprotected which can be observed inconsistently
     causing state corruption.

   - A trivial compile fix for CONFIG_CGROUPS=n"

* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched: Fix pick_next_task() vs 'change' pattern race
  sched/core: Fix compilation error when cgroup not selected

Pull perf tooling fixes from Thomas Gleixner:

 - Fix the time sorting algorithm which was broken due to truncation of
   big numbers

 - Fix the python script generator fail caused by a broken tracepoint
   array iterator

* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf tools: Fix time sorting
  perf tools: Remove unused trace_find_next_event()
  perf scripting engines: Iterate on tep event arrays directly

Pull irq fixlet from Thomas Gleixner:
 "A trivial fix for a kernel doc regression where an argument change was
  not reflected in the documentation"

* 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  irq/irqdomain: Update __irq_domain_alloc_fwnode() function documentation

Pull stacktrace fix from Thomas Gleixner:
 "A small fix for a stacktrace regression.

  Saving a stacktrace for a foreign task skipped an extra entry which
  makes e.g. the output of /proc/$PID/stack incomplete"

* 'core-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  stacktrace: Don't skip first entry on noncurrent tasks

Pull cifs fix from Steve French:
 "Small fix for an smb3 reconnect bug (also marked for stable)"

* tag '5.4-rc7-smb3-fix' of git://git.samba.org/sfrench/cifs-2.6:
  SMB3: Fix persistent handles reconnect

config option GENERIC_IO was removed but still selected by lib/kconfig
This patch finish the cleaning.

Fixes: 9de8da47 ("kconfig: kill off GENERIC_IO option")
Acked-by: Rob Herring <robh@kernel.org>
Signed-off-by: Corentin Labbe <clabbe@baylibre.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Pull pin control fixes from Linus Walleij:

 - Fix glitch risks in the Intel GPIO

 - Fix the Intel Cherryview valid irq mask calculation.

 - Allocate the Intel Cherryview irqchip dynamically.

 - Fix the valid mask init sequency on the ST STMFX driver.

* tag 'pinctrl-v5.4-3' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-pinctrl:
  pinctrl: stmfx: fix valid_mask init sequence
  pinctrl: cherryview: Allocate IRQ chip dynamic
  pinctrl: cherryview: Fix irq_valid_mask calculation
  pinctrl: intel: Avoid potential glitches if pin is in GPIO mode

Pull btrfs fixes from David Sterba:
 "A few regressions and fixes for stable.

  Regressions:

   - fix a race leading to metadata space leak after task received a
     signal

   - un-deprecate 2 ioctls, marked as deprecated by mistake

  Fixes:

   - fix limit check for number of devices during chunk allocation

   - fix a race due to double evaluation of i_size_read inside max()
     macro, can cause a crash

   - remove wrong device id check in tree-checker"

* tag 'for-5.4-rc6-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
  btrfs: un-deprecate ioctls START_SYNC and WAIT_SYNC
  btrfs: save i_size to avoid double evaluation of i_size_read in compress_file_range
  Btrfs: fix race leading to metadata space leak after task received signal
  btrfs: tree-checker: Fix wrong check on max devid
  btrfs: Consider system chunk array size for new SYSTEM chunks

Pull watchdog fixes from Wim Van Sebroeck:

 - cpwd: fix build regression

 - pm8916_wdt: fix pretimeout registration flow

 - meson: Fix the wrong value of left time

 - imx_sc_wdt: Pretimeout should follow SCU firmware format

 - bd70528: Add MODULE_ALIAS to allow module auto loading

* tag 'linux-watchdog-5.4-rc7' of git://www.linux-watchdog.org/linux-watchdog:
  watchdog: bd70528: Add MODULE_ALIAS to allow module auto loading
  watchdog: imx_sc_wdt: Pretimeout should follow SCU firmware format
  watchdog: meson: Fix the wrong value of left time
  watchdog: pm8916_wdt: fix pretimeout registration flow
  watchdog: cpwd: fix build regression

Pull networking fixes from David Miller:

 1) BPF sample build fixes from Björn Töpel

 2) Fix powerpc bpf tail call implementation, from Eric Dumazet.

 3) DCCP leaks jiffies on the wire, fix also from Eric Dumazet.

 4) Fix crash in ebtables when using dnat target, from Florian Westphal.

 5) Fix port disable handling whne removing bcm_sf2 driver, from Florian
    Fainelli.

 6) Fix kTLS sk_msg trim on fallback to copy mode, from Jakub Kicinski.

 7) Various KCSAN fixes all over the networking, from Eric Dumazet.

 8) Memory leaks in mlx5 driver, from Alex Vesker.

 9) SMC interface refcounting fix, from Ursula Braun.

10) TSO descriptor handling fixes in stmmac driver, from Jose Abreu.

11) Add a TX lock to synchonize the kTLS TX path properly with crypto
    operations. From Jakub Kicinski.

12) Sock refcount during shutdown fix in vsock/virtio code, from Stefano
    Garzarella.

13) Infinite loop in Intel ice driver, from Colin Ian King.

* git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (108 commits)
  ixgbe: need_wakeup flag might not be set for Tx
  i40e: need_wakeup flag might not be set for Tx
  igb/igc: use ktime accessors for skb->tstamp
  i40e: Fix for ethtool -m issue on X722 NIC
  iavf: initialize ITRN registers with correct values
  ice: fix potential infinite loop because loop counter being too small
  qede: fix NULL pointer deref in __qede_remove()
  net: fix data-race in neigh_event_send()
  vsock/virtio: fix sock refcnt holding during the shutdown
  net: ethernet: octeon_mgmt: Account for second possible VLAN header
  mac80211: fix station inactive_time shortly after boot
  net/fq_impl: Switch to kvmalloc() for memory allocation
  mac80211: fix ieee80211_txq_setup_flows() failure path
  ipv4: Fix table id reference in fib_sync_down_addr
  ipv6: fixes rt6_probe() and fib6_nh->last_probe init
  net: hns: Fix the stray netpoll locks causing deadlock in NAPI path
  net: usb: qmi_wwan: add support for DW5821e with eSIM support
  CDC-NCM: handle incomplete transfer of MTU
  nfc: netlink: fix double device reference drop
  NFC: st21nfca: fix double free
  ...

Pull block fixes from Jens Axboe:

 - Two NVMe device removal crash fixes, and a compat fixup for for an
   ioctl that was introduced in this release (Anton, Charles, Max - via
   Keith)

 - Missing error path mutex unlock for drbd (Dan)

 - cgroup writeback fixup on dead memcg (Tejun)

 - blkcg online stats print fix (Tejun)

* tag 'for-linus-2019-11-08' of git://git.kernel.dk/linux-block:
  cgroup,writeback: don't switch wbs immediately on dead wbs if the memcg is dead
  block: drbd: remove a stray unlock in __drbd_send_protocol()
  blkcg: make blkcg_print_stat() print stats only for online blkgs
  nvme: change nvme_passthru_cmd64 to explicitly mark rsvd
  nvme-multipath: fix crash in nvme_mpath_clear_ctrl_paths
  nvme-rdma: fix a segmentation fault during module unload

Jeff Kirsher says:

====================
Intel Wired LAN Driver Fixes 2019-11-08

This series contains fixes to igb, igc, ixgbe, i40e, iavf and ice
drivers.

Colin Ian King fixes a potentially wrap-around counter in a for-loop.

Nick fixes the default ITR values for the iavf driver to 50 usecs
interval.

Arkadiusz fixes 'ethtool -m' for X722 devices where the correct value
cannot be obtained from the firmware, so add X722 to the check to ensure
the wrong value is not returned.

Jake fixes igb and igc drivers in their implementation of launch time
support by declaring skb->tstamp value as ktime_t instead of s64.

Magnus fixes ixgbe and i40e where the need_wakeup flag for transmit may
not be set for AF_XDP sockets that are only used to send packets.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

The need_wakeup flag for Tx might not be set for AF_XDP sockets that
are only used to send packets. This happens if there is at least one
outstanding packet that has not been completed by the hardware and we
get that corresponding completion (which will not generate an
interrupt since interrupts are disabled in the napi poll loop) between
the time we stopped processing the Tx completions and interrupts are
enabled again. In this case, the need_wakeup flag will have been
cleared at the end of the Tx completion processing as we believe we
will get an interrupt from the outstanding completion at a later point
in time. But if this completion interrupt occurs before interrupts
are enable, we lose it and should at that point really have set the
need_wakeup flag since there are no more outstanding completions that
can generate an interrupt to continue the processing. When this
happens, user space will see a Tx queue need_wakeup of 0 and skip
issuing a syscall, which means will never get into the Tx processing
again and we have a deadlock.

This patch introduces a quick fix for this issue by just setting the
need_wakeup flag for Tx to 1 all the time. I am working on a proper
fix for this that will toggle the flag appropriately, but it is more
challenging than I anticipated and I am afraid that this patch will
not be completed before the merge window closes, therefore this easier
fix for now. This fix has a negative performance impact in the range
of 0% to 4%. Towards the higher end of the scale if you have driver
and application on the same core and issue a lot of packets, and
towards no negative impact if you use two cores, lower transmission
speeds and/or a workload that also receives packets.
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

The need_wakeup flag for Tx might not be set for AF_XDP sockets that
are only used to send packets. This happens if there is at least one
outstanding packet that has not been completed by the hardware and we
get that corresponding completion (which will not generate an
interrupt since interrupts are disabled in the napi poll loop) between
the time we stopped processing the Tx completions and interrupts are
enabled again. In this case, the need_wakeup flag will have been
cleared at the end of the Tx completion processing as we believe we
will get an interrupt from the outstanding completion at a later point
in time. But if this completion interrupt occurs before interrupts
are enable, we lose it and should at that point really have set the
need_wakeup flag since there are no more outstanding completions that
can generate an interrupt to continue the processing. When this
happens, user space will see a Tx queue need_wakeup of 0 and skip
issuing a syscall, which means will never get into the Tx processing
again and we have a deadlock.

This patch introduces a quick fix for this issue by just setting the
need_wakeup flag for Tx to 1 all the time. I am working on a proper
fix for this that will toggle the flag appropriately, but it is more
challenging than I anticipated and I am afraid that this patch will
not be completed before the merge window closes, therefore this easier
fix for now. This fix has a negative performance impact in the range
of 0% to 4%. Towards the higher end of the scale if you have driver
and application on the same core and issue a lot of packets, and
towards no negative impact if you use two cores, lower transmission
speeds and/or a workload that also receives packets.
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>