Revert commit 3510fac4 (cpufreq: postfix policy directory with the
first CPU in related_cpus).

Earlier, the policy->kobj was added to the kobject core, before ->init()
callback was called for the cpufreq drivers. Which allowed those drivers
to add or remove, driver dependent, sysfs files/directories to the same
kobj from their ->init() and ->exit() callbacks.

That isn't possible anymore after commit 3510fac4.

Now, there is no other clean alternative that people can adopt.

Its better to revert the earlier commit to allow cpufreq drivers to
create/remove sysfs files from ->init() and ->exit() callbacks.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

Setting a new CPU frequency and reading the current request value
in the ACPI cpufreq driver involves each at least two switch
instructions (there's more if the policy is shared).  One of
them is present in drv_read/write() that prepares a command
structure and the other happens in subsequent do_drv_read/write()
when that structure is interpreted.  However, all of those switches
may be avoided by using function pointers.

To that end, add two function pointers to struct acpi_cpufreq_data
to represent read and write operations on the frequency register
and set them up during policy intitialization to point to the pair
of routines suitable for the given processor (Intel/AMD MSR access
or I/O port access).  Then, use those pointers in do_drv_read/write()
and modify drv_read/write() to prepare the command structure for
them without any checks.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

Unregister the notifiers if cpufreq_driver_register() fails in
powernv_cpufreq_init(). Re-arrange the unregistration and cleanup routines
in powernv_cpufreq_exit() to free all the resources after the driver
has unregistered.
Signed-off-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

Disable HWP Interrupt notification before enabling HWP. Since we don't
have HWP interrupt handling for possible performance interrupts, there
is not much use of enabling HWP interrupts.
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

If the processor supports HWP, enable it by default without checking
for the cpu model. This will allow to enable HWP in all supported
processors without driver change.
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

That macro uses an internal static inline function that is first
totally unnecessary and second hard to read, so simplify it and
get rid of that monster.

No functional changes.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>

The "out" label at the final return statement in acpi_cpufreq_target()
is totally pointless, so drop them and modify the code to return the
right values immediately instead of jumping to it.

No functional changes.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>

Drop a pointless label at a return statement from
__cpufreq_driver_target() and rearrange that function
to reduce the indentation level.

No intentional functional changes.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>

The intel-pstate driver is using intel_pstate_hwp_set() from two
separate paths, i.e. ->set_policy() callback and sysfs update path for
the files present in /sys/devices/system/cpu/intel_pstate/ directory.

While an update to the sysfs path applies to all the CPUs being managed
by the driver (which essentially means all the online CPUs), the update
via the ->set_policy() callback applies to a smaller group of CPUs
managed by the policy for which ->set_policy() is called.

And so, intel_pstate_hwp_set() should update frequencies of only the
CPUs that are part of policy->cpus mask, while it is called from
->set_policy() callback.

In order to do that, add a parameter (cpumask) to intel_pstate_hwp_set()
and apply the frequency changes only to the concerned CPUs.

For ->set_policy() path, we are only concerned about policy->cpus, and
so policy->rwsem lock taken by the core prior to calling ->set_policy()
is enough to take care of any races. The larger lock acquired by
get_online_cpus() is required only for the updates to sysfs files.

Add another routine, intel_pstate_hwp_set_online_cpus(), and call it
from the sysfs update paths.

This also fixes a lockdep reported recently, where policy->rwsem and
get_online_cpus() could have been acquired in any order causing an ABBA
deadlock. The sequence of events leading to that was:

intel_pstate_init(...)
	...cpufreq_online(...)
		down_write(&policy->rwsem); // Locks policy->rwsem
		...
		cpufreq_init_policy(policy);
			...intel_pstate_hwp_set();
				get_online_cpus(); // Temporarily locks cpu_hotplug.lock
		...
		up_write(&policy->rwsem);

pm_suspend(...)
	...disable_nonboot_cpus()
		_cpu_down()
			cpu_hotplug_begin(); // Locks cpu_hotplug.lock
			__cpu_notify(CPU_DOWN_PREPARE, ...);
				...cpufreq_offline_prepare();
					down_write(&policy->rwsem); // Locks policy->rwsem
Reported-and-tested-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Acked-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

Signed-off-by: Eric Biggers <ebiggers3@gmail.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

The comment has been incorrect since commit 4dea5806
("cpufreq: return EEXIST instead of EBUSY for second registering").
Signed-off-by: Eric Biggers <ebiggers3@gmail.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

This just swaps a colon for a quote in the intel_pstate documentation.
Signed-off-by: Felipe Franciosi <felipe@nutanix.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

The show() and store() routines in the cpufreq core don't need to
check if the struct freq_attr they want to use really provides the
callbacks they need as expected (if that's not the case, it means
a bug in the code anyway), so change them to avoid doing that.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>

OPP layer manages it now and cpufreq-dt driver doesn't need it. But, we
still need to check for availability of resources for deferred probing.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

Its already done by core and we don't need to get it anymore.  And so,
we don't need to get of node in cpufreq_init() anymore, move that to
find_supply_name() instead.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

OPP core supports frequency/voltage changes based on the target
frequency now, use that instead of open coding the same in cpufreq-dt
driver.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

OPP layer has all the information now to calculate transition latency
(clock_latency + voltage_latency). Lets reuse the OPP layer helper
dev_pm_opp_get_max_transition_latency() instead of open coding the same
in cpufreq-dt driver.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

The core already have a valid regulator set for the device opp and the
unsupported OPPs are already disabled by the core. There is no need to
repeat that in the user drivers, get rid of it.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

OPP core can handle the regulators by itself, and but it needs to know
the name of the regulator to fetch. Add support for that.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

"clock-latency" is handled by OPP layer for all bindings and so there is
no need to make special calls for V1 bindings. Use
dev_pm_opp_get_max_clock_latency() for both the cases.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

That's the real purpose of this field, i.e. to take special care of old
OPP V1 bindings. Lets name it accordingly, so that it can be used
elsewhere.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

We have the device structure available now, lets use it for better print
messages.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

This adds a routine, dev_pm_opp_set_rate(), responsible for configuring
power-supply and clock source for an OPP.

The OPP is found by matching against the target_freq passed to the
routine. This shall replace similar code present in most of the OPP
users and help simplify them a lot.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

OPP core has got almost everything now to manage device's OPP
transitions, the only thing left is device's clk. Get that as well.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

V2 bindings have better support for clock-latency and voltage-tolerance
and doesn't need special care. To use callbacks, like
dev_pm_opp_get_max_{transition|volt}_latency(), irrespective of the
bindings, the core needs to know clock-latency/voltage-tolerance for the
earlier bindings.

This patch reads clock-latency/voltage-tolerance from the device node,
irrespective of the bindings (to keep it simple) and use them only for
V1 bindings.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

In few use cases (like: cpufreq), it is desired to get the maximum
latency for changing OPPs. Add support for that.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

In few use cases (like: cpufreq), it is desired to get the maximum
voltage latency for changing OPPs. Add support for that.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

Disable any OPPs where the connected regulator isn't able to provide the
specified voltage.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

This allows the OPP core to request/free the regulator resource,
attached to a device OPP. The regulator device is fetched using the name
provided by the driver, while calling: dev_pm_opp_set_regulator().

This will work for both OPP-v1 and v2 bindings.

This is a preliminary step for moving the OPP switching logic into the
OPP core.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

Pull ARM SoC fixes from Olof Johansson:
 "The first real batch of fixes for this release cycle, so there are a
  few more than usual.

  Most of these are fixes and tweaks to board support (DT bugfixes,
  etc).  I've also picked up a couple of small cleanups that seemed
  innocent enough that there was little reason to wait (const/
  __initconst and Kconfig deps).

  Quite a bit of the changes on OMAP were due to fixes to no longer
  write to rodata from assembly when ARM_KERNMEM_PERMS was enabled, but
  there were also other fixes.

  Kirkwood had a bunch of gpio fixes for some boards.  OMAP had RTC
  fixes on OMAP5, and Nomadik had changes to MMC parameters in DT.

  All in all, mostly the usual mix of various fixes"

* tag 'armsoc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc: (46 commits)
  ARM: multi_v7_defconfig: enable DW_WATCHDOG
  ARM: nomadik: fix up SD/MMC DT settings
  ARM64: tegra: Add chosen node for tegra132 norrin
  ARM: realview: use "depends on" instead of "if" after prompt
  ARM: tango: use "depends on" instead of "if" after prompt
  ARM: tango: use const and __initconst for smp_operations
  ARM: realview: use const and __initconst for smp_operations
  bus: uniphier-system-bus: revive tristate prompt
  arm64: dts: Add missing DMA Abort interrupt to Juno
  bus: vexpress-config: Add missing of_node_put
  ARM: dts: am57xx: sbc-am57x: correct Eth PHY settings
  ARM: dts: am57xx: cl-som-am57x: fix CPSW EMAC pinmux
  ARM: dts: am57xx: sbc-am57x: fix UART3 pinmux
  ARM: dts: am57xx: cl-som-am57x: update SPI Flash frequency
  ARM: dts: am57xx: cl-som-am57x: set HOST mode for USB2
  ARM: dts: am57xx: sbc-am57x: fix SB-SOM EEPROM I2C address
  ARM: dts: LogicPD Torpedo: Revert Duplicative Entries
  ARM: dts: am437x: pixcir_tangoc: use correct flags for irq types
  ARM: dts: am4372: fix irq type for arm twd and global timer
  ARM: dts: at91: sama5d4 xplained: fix phy0 IRQ type
  ...

Pull mailbox fixes from Jassi Brar:

 - fix getting element from the pcc-channels array by simply indexing
   into it

 - prevent building mailbox-test driver for archs that don't have IOMEM

* 'mailbox-devel' of git://git.linaro.org/landing-teams/working/fujitsu/integration:
  mailbox: Fix dependencies for !HAS_IOMEM archs
  mailbox: pcc: fix channel calculation in get_pcc_channel()

Pull USB fixes from Greg KH:
 "Here are some USB fixes for 4.5-rc3.

  The usual, xhci fixes for reported issues, combined with some small
  gadget driver fixes, and a MAINTAINERS file update.  All have been in
  linux-next with no reported issues"

* tag 'usb-4.5-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb:
  xhci: harden xhci_find_next_ext_cap against device removal
  xhci: Fix list corruption in urb dequeue at host removal
  usb: host: xhci-plat: fix NULL pointer in probe for device tree case
  usb: xhci-mtk: fix AHB bus hang up caused by roothubs polling
  usb: xhci-mtk: fix bpkts value of LS/HS periodic eps not behind TT
  usb: xhci: apply XHCI_PME_STUCK_QUIRK to Intel Broxton-M platforms
  usb: xhci: set SSIC port unused only if xhci_suspend succeeds
  usb: xhci: add a quirk bit for ssic port unused
  usb: xhci: handle both SSIC ports in PME stuck quirk
  usb: dwc3: gadget: set the OTG flag in dwc3 gadget driver.
  Revert "xhci: don't finish a TD if we get a short-transfer event mid TD"
  MAINTAINERS: fix my email address
  usb: dwc2: Fix probe problem on bcm2835
  Revert "usb: dwc2: Move reset into dwc2_get_hwparams()"
  usb: musb: ux500: Fix NULL pointer dereference at system PM
  usb: phy: mxs: declare variable with initialized value
  usb: phy: msm: fix error handling in probe.

Pull staging and IIO driver fixes from Greg KH:
 "Here are some IIO and staging driver fixes for 4.5-rc3.

  All of them, except one, are for IIO drivers, and one is for a speakup
  driver fix caused by some earlier patches, to resolve a reported build
  failure"

* tag 'staging-4.5-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging:
  Staging: speakup: Fix allyesconfig build on mn10300
  iio: dht11: Use boottime
  iio: ade7753: avoid uninitialized data
  iio: pressure: mpl115: fix temperature offset sign
  iio: imu: Fix dependencies for !HAS_IOMEM archs
  staging: iio: Fix dependencies for !HAS_IOMEM archs
  iio: adc: Fix dependencies for !HAS_IOMEM archs
  iio: inkern: fix a NULL dereference on error
  iio:adc:ti_am335x_adc Fix buffered mode by identifying as software buffer.
  iio: light: acpi-als: Report data as processed
  iio: dac: mcp4725: set iio name property in sysfs
  iio: add HAS_IOMEM dependency to VF610_ADC
  iio: add IIO_TRIGGER dependency to STK8BA50
  iio: proximity: lidar: correct return value
  iio-light: Use a signed return type for ltr501_match_samp_freq()

Merge fixes from Andrew Morton:
 "22 fixes"

* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (22 commits)
  epoll: restrict EPOLLEXCLUSIVE to POLLIN and POLLOUT
  radix-tree: fix oops after radix_tree_iter_retry
  MAINTAINERS: trim the file triggers for ABI/API
  dax: dirty inode only if required
  thp: make deferred_split_scan() work again
  mm: replace vma_lock_anon_vma with anon_vma_lock_read/write
  ocfs2/dlm: clear refmap bit of recovery lock while doing local recovery cleanup
  um: asm/page.h: remove the pte_high member from struct pte_t
  mm, hugetlb: don't require CMA for runtime gigantic pages
  mm/hugetlb: fix gigantic page initialization/allocation
  mm: downgrade VM_BUG in isolate_lru_page() to warning
  mempolicy: do not try to queue pages from !vma_migratable()
  mm, vmstat: fix wrong WQ sleep when memory reclaim doesn't make any progress
  vmstat: make vmstat_update deferrable
  mm, vmstat: make quiet_vmstat lighter
  mm/Kconfig: correct description of DEFERRED_STRUCT_PAGE_INIT
  memblock: don't mark memblock_phys_mem_size() as __init
  dump_stack: avoid potential deadlocks
  mm: validate_mm browse_rb SMP race condition
  m32r: fix build failure due to SMP and MMU
  ...

Pull Ceph fixes from Sage Weil:
 "We have a few wire protocol compatibility fixes, ports of a few recent
  CRUSH mapping changes, and a couple error path fixes"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client:
  libceph: MOSDOpReply v7 encoding
  libceph: advertise support for TUNABLES5
  crush: decode and initialize chooseleaf_stable
  crush: add chooseleaf_stable tunable
  crush: ensure take bucket value is valid
  crush: ensure bucket id is valid before indexing buckets array
  ceph: fix snap context leak in error path
  ceph: checking for IS_ERR instead of NULL

Pull drm fixes from Dave Airlie:
 "Fixes all over the place:

   - amdkfd: two static checker fixes
   - mst: a bunch of static checker and spec/hw interaction fixes
   - amdgpu: fix Iceland hw properly, and some fiji bugs, along with
     some write-combining fixes.
   - exynos: some regression fixes
   - adv7511: fix some EDID reading issues"

* 'drm-fixes' of git://people.freedesktop.org/~airlied/linux: (38 commits)
  drm/dp/mst: deallocate payload on port destruction
  drm/dp/mst: Reverse order of MST enable and clearing VC payload table.
  drm/dp/mst: move GUID storage from mgr, port to only mst branch
  drm/dp/mst: change MST detection scheme
  drm/dp/mst: Calculate MST PBN with 31.32 fixed point
  drm: Add drm_fixp_from_fraction and drm_fixp2int_ceil
  drm/mst: Add range check for max_payloads during init
  drm/mst: Don't ignore the MST PBN self-test result
  drm: fix missing reference counting decrease
  drm/amdgpu: disable uvd and vce clockgating on Fiji
  drm/amdgpu: remove exp hardware support from iceland
  drm/amdgpu: load MEC ucode manually on iceland
  drm/amdgpu: don't load MEC2 on topaz
  drm/amdgpu: drop topaz support from gmc8 module
  drm/amdgpu: pull topaz gmc bits into gmc_v7
  drm/amdgpu: The VI specific EXE bit should only apply to GMC v8.0 above
  drm/amdgpu: iceland use CI based MC IP
  drm/amdgpu: move gmc7 support out of CIK dependency
  drm/amdgpu/gfx7: enable cp inst/reg error interrupts
  drm/amdgpu/gfx8: enable cp inst/reg error interrupts
  ...

Pull power management and ACPI fixes from Rafael Wysocki:
 "These are: a fix for a recently introduced false-positive warnings
  about PM domain pointers being changed inappropriately (harmless but
  annoying), an MCH size workaround quirk for one more platform, a
  compiler warning fix (generic power domains framework), an ACPI LPSS
  (Intel SoCs) driver fixup and a cleanup of the ACPI CPPC core code.

  Specifics:

   - PM core fix to avoid false-positive warnings generated when the
     pm_domain field is cleared for a device that appears to be bound to
     a driver (Rafael Wysocki).

   - New MCH size workaround quirk for Intel Haswell-ULT (Josh Boyer).

   - Fix for an "unused function" compiler warning in the generic power
     domains framework (Ulf Hansson).

   - Fixup for the ACPI driver for Intel SoCs (acpi-lpss) to set the PM
     domain pointer of a device properly in one place that was
     overlooked by a recent PM core update (Andy Shevchenko).

   - Removal of a redundant function declaration in the ACPI CPPC core
     code (Timur Tabi)"

* tag 'pm+acpi-4.5-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  PM: Avoid false-positive warnings in dev_pm_domain_set()
  PM / Domains: Silence compiler warning for an unused function
  ACPI / CPPC: remove redundant mbox_send_message() declaration
  ACPI / LPSS: set PM domain via helper setter
  PNP: Add Haswell-ULT to Intel MCH size workaround

In the current implementation of the EPOLLEXCLUSIVE flag (added for
4.5-rc1), if epoll waiters create different POLL* sets and register them
as exclusive against the same target fd, the current implementation will
stop waking any further waiters once it finds the first idle waiter.
This means that waiters could miss wakeups in certain cases.

For example, when we wake up a pipe for reading we do:
wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM); So if
one epoll set or epfd is added to pipe p with POLLIN and a second set
epfd2 is added to pipe p with POLLRDNORM, only epfd may receive the
wakeup since the current implementation will stop after it finds any
intersection of events with a waiter that is blocked in epoll_wait().

We could potentially address this by requiring all epoll waiters that
are added to p be required to pass the same set of POLL* events.  IE the
first EPOLL_CTL_ADD that passes EPOLLEXCLUSIVE establishes the set POLL*
flags to be used by any other epfds that are added as EPOLLEXCLUSIVE.
However, I think it might be somewhat confusing interface as we would
have to reference count the number of users for that set, and so
userspace would have to keep track of that count, or we would need a
more involved interface.  It also adds some shared state that we'd have
store somewhere.  I don't think anybody will want to bloat
__wait_queue_head for this.

I think what we could do instead, is to simply restrict EPOLLEXCLUSIVE
such that it can only be specified with EPOLLIN and/or EPOLLOUT.  So
that way if the wakeup includes 'POLLIN' and not 'POLLOUT', we can stop
once we hit the first idle waiter that specifies the EPOLLIN bit, since
any remaining waiters that only have 'POLLOUT' set wouldn't need to be
woken.  Likewise, we can do the same thing if 'POLLOUT' is in the wakeup
bit set and not 'POLLIN'.  If both 'POLLOUT' and 'POLLIN' are set in the
wake bit set (there is at least one example of this I saw in fs/pipe.c),
then we just wake the entire exclusive list.  Having both 'POLLOUT' and
'POLLIN' both set should not be on any performance critical path, so I
think that's ok (in fs/pipe.c its in pipe_release()).  We also continue
to include EPOLLERR and EPOLLHUP by default in any exclusive set.  Thus,
the user can specify EPOLLERR and/or EPOLLHUP but is not required to do
so.

Since epoll waiters may be interested in other events as well besides
EPOLLIN, EPOLLOUT, EPOLLERR and EPOLLHUP, these can still be added by
doing a 'dup' call on the target fd and adding that as one normally
would with EPOLL_CTL_ADD.  Since I think that the POLLIN and POLLOUT
events are what we are interest in balancing, I think that the 'dup'
thing could perhaps be added to only one of the waiter threads.
However, I think that EPOLLIN, EPOLLOUT, EPOLLERR and EPOLLHUP should be
sufficient for the majority of use-cases.

Since EPOLLEXCLUSIVE is intended to be used with a target fd shared
among multiple epfds, where between 1 and n of the epfds may receive an
event, it does not satisfy the semantics of EPOLLONESHOT where only 1
epfd would get an event.  Thus, it is not allowed to be specified in
conjunction with EPOLLEXCLUSIVE.

EPOLL_CTL_MOD is also not allowed if the fd was previously added as
EPOLLEXCLUSIVE.  It seems with the limited number of flags to not be as
interesting, but this could be relaxed at some further point.
Signed-off-by: Jason Baron <jbaron@akamai.com>
Tested-by: Madars Vitolins <m@silodev.com>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Al Viro <viro@ftp.linux.org.uk>
Cc: Eric Wong <normalperson@yhbt.net>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Hagen Paul Pfeifer <hagen@jauu.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>