Though we know the hw is limited to keeping stolen memory inside the
first 4GiB, it is clearer to the reader that we are handling physical
address if we use phys_addr_t to refer to the base of stolen memory.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NTvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170106152013.24684-2-chris@chris-wilson.co.uk

In order to silence sparse:

../drivers/gpu/drm/i915/i915_gpu_error.c:200:39: warning: Using plain integer as NULL pointer

add a helper to check whether we have sse4.1 and that the desired
alignment is valid for acceleration.

v2: Explain the macros and split the two use cases between
i915_has_memcpy_from_wc() and i915_can_memcpy_from_wc().
Reported-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NTvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170106152013.24684-1-chris@chris-wilson.co.uk

If we skip before banning, we have an inconsistent interface between
execbuf still queueing valid request but those requests already queued
being cancelled. If we only cancel the pending requests once we stop
accepting new requests, the user interface is more consistent.
Reported-by: NTvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Fixes: 821ed7df ("drm/i915: Update reset path to fix incomplete requests")
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Cc: <stable@vger.kernel.org> # v4.9+
Reviewed-by: NTvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170105170059.344-1-chris@chris-wilson.co.uk

Now that we have split out a header file for simple macros (that maybe
we can promote into a core header), move a few macros across from
i915_drv.h
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20170105164148.26875-1-chris@chris-wilson.co.ukReviewed-by: NMatthew Auld <matthew.auld@intel.com>
Reviewed-by: NTvrtko Ursulin <tvrtko.ursulin@intel.com>

Missed when rebasing patches, I failed to set ret to zero before
starting the unbind loop (which depends upon ret being zero).
Reported-by: NMatthew Auld <matthew.william.auld@gmail.com>
Fixes: 9332f3b1 ("drm/i915: Combine loops within i915_gem_evict_something")
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.william.auld@gmail.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170105155940.10033-1-chris@chris-wilson.co.ukReviewed-by: NMatthew Auld <matthew.william.auld@gmail.com>
Cc: <stable@vger.kernel.org> # v4.9+

The GuC uses a special mapping for the upper end of the Global GTT,
similar to the way it uses a special mapping for the lower end, so
exclude it from our drm_mm to prevent us using it.

v2: Rename to reflect that it is unmappable similar to the region at the
bottom of the GGTT, and couple it into the assertion that we don't feed
unmappable addresses to the GuC.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Arkadiusz Hiler <arkadiusz.hiler@intel.com>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: NArkadiusz Hiler <arkadiusz.hiler@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170105153023.30575-5-chris@chris-wilson.co.uk

In order to defeat some circular dependencies between headers to allow use
of e.g. range_overflows() in a header, move the simple independent macros
into their own header.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NTvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170105153023.30575-4-chris@chris-wilson.co.uk

Empirically we restart following a GPU reset more successfully if we call
lrc_init_hws() (which contains a posting read) last. (The failure mode
that was observed was that breadcrumb writes into the HWS from the
recovered requests went astray leading to the context-switch maintaining
forward progress, but the requests not being retired/completed.)

For clarity, lrc_init_hws() is inlined (and the unused function then
removed).
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: NTvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170105153023.30575-3-chris@chris-wilson.co.uk

In order to convince static analyzers that the allocation function
returns an error or sets ce->state, assert that it is set afterwards.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NTvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170105153023.30575-2-chris@chris-wilson.co.uk

During i915_gem_timeline_fini(), assert that all the timeline's request
are completed and removed from the timeline.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NTvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170105153023.30575-1-chris@chris-wilson.co.uk

The fence registers are clobbered by a GPU reset. If there is concurrent
user access to a fenced region via a GTT mmaping, the access will not be
fenced during the reset (until we restore the fences afterwards). In order
to prevent invalid access during the reset, before we clobber the fences
first we must invalidate the GTT mmapings. Access to the mmap will then
be forced to fault in the page, and in handling the fault, i915_gem_fault()
will take the struct_mutex and wait upon the reset to complete.

v2: Fix up commentary.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99274
Testcase: igt/gem_mmap_gtt/hang
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20170104145110.1486-1-chris@chris-wilson.co.ukReviewed-by: NTvrtko Ursulin <tvrtko.ursulin@intel.com>

Gen9+ platforms have been seeing a lot of screen flickerings and
underruns, so I never felt comfortable in enabling FBC on these
platforms since I didn't want to throw yet another feature on top of
the already complex problem. We now have code that automatically
disables FBC if we ever get an underrun, and the screen flickerings
seem to be mostly gone, so it may be a good time to try to finally
enable FBC by default on the newer platforms.

Besides, BDW FBC has been working fine over the year, which gives me a
little more confidence now.

For a little more information, please refer to commit a98ee793
("drm/i915/fbc: enable FBC by default on HSW and BDW").

v2: Enable not only on SKL, but for everything new (Daniel).
v3: Rebase after the intel_sanitize_fbc_option() change.
v4: New rebase after 8 months, drop expired R-B tags.
Signed-off-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1482495839-27041-1-git-send-email-paulo.r.zanoni@intel.com

Back in 2014, commit fb7023e0 ("drm/i915: BDW: Adding Reserved PCI
IDs.") added the reserved PCI IDs in order to try to make sure we had
working drivers in case we ever released products using these IDs
(since we had instances of this type of problem in the past). The
problem is that the patch only touched the macros used by
early-quirks.c and by the user space components that rely on
i915_pciids.h, it didn't touch the macros used by i915_pci.c. So we
correctly handled the stolen memory for these theoretical IDs, but we
didn't actually drive the devices from i915.ko.

So this patch fixes the original commit by actually making i915.ko
drive these IDs, which was the goal. There's no information on what
would be the GT count on these IDs, so we just go with the safer
intel_broadwell_info, at the risk of ignoring a possibly inexistent
BSD2_RING.

I did some checking, and it seems that these IDs are driven by
intel-gpu-tools, xf86-video-intel and libdrm (since they contain old
copies of i915_pciids.h), but they are not checked by mesa.

The alternative to this patch would be to just assume we're actually
never going to use these IDs, and then remove them from our ID lists
and make sure our user space components sync the latest i915_pciids.h
copy. I'm fine with either approaches, as long as we make sure that
every component tries to drive the same list of PCI IDs.

Fixes: fb7023e0 ("drm/i915: BDW: Adding Reserved PCI IDs.")
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Ben Widawsky <ben@bwidawsk.net>
Cc: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1483473860-17644-3-git-send-email-paulo.r.zanoni@intel.com

Commit 8d9c20e1 ("drm/i915: Remove .is_mobile field from platform
struct") removed mobile vs desktop differences for HSW+, but forgot
the Broadwell reserved IDs, so do it now.

It's interesting to notice that these IDs are used by early-quirks.c
but are *not* used by i915_pci.c.

Cc: Carlos Santa <carlos.santa@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1483473860-17644-2-git-send-email-paulo.r.zanoni@intel.com

Remove duplicated IDs from the list. Currently, this definition is
only used by early-quirks.c. From my understanding of the code, having
duplicated IDs shouldn't be causing any bugs.

Fixes: 8d9c20e1 ("drm/i915: Remove .is_mobile field from platform struct")
Cc: Carlos Santa <carlos.santa@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1483473860-17644-1-git-send-email-paulo.r.zanoni@intel.com

Merge tag 'drm-misc-next-2016-12-30' of git://anongit.freedesktop.org/git/drm-misc into drm-intel-next-queued

Directly merge drm-misc into drm-intel since Dave is on vacation and
we need the various drm-misc patches (fb format rework, drm mm fixes,
selftest framework and others). Also pulled back -rc2 in first to
resync with drm-intel-fixes and make sure I can reuse the exact rerere
solutions from drm-tip for safety, and because I'm lazy.
Signed-off-by: NDaniel Vetter <daniel.vetter@intel.com>

Backmerge Linux 4.10-rc2 to resync with our -fixes cherry-picks. I've
done the backmerge directly because Dave is on vacation.
Signed-off-by: NDaniel Vetter <daniel.vetter@intel.com>

The code was moved, but the comment not updated. It confused me.

Fixes: 7f4c6284 ("drm/i915: Assign hwmode after encoder state readout")
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Patrik Jakobsson <patrik.jakobsson@linux.intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NMaarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161219082423.27798-6-daniel.vetter@ffwll.ch

No functional changes. Apparently spec has been changed
the valid table showing 0x192A as Server GT4
while 0x193A is Server GT4e.

Libdrm and Mesa already have this right. So let's fix the ref here.

Cc: Ben Widawsky <benjamin.widawsky@intel.com>
Signed-off-by: NRodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: NBen Widawsky <benjamin.widawsky@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1483471672-10450-1-git-send-email-rodrigo.vivi@intel.com

After commit 1c74eeaf ("drm/i915: Move number of scalers initialization to
runtime init"), scalers are not initialized properly for skl and glk
since num_scalers is left as 0 for those platforms.

Fixes: 1c74eeaf ("drm/i915: Move number of scalers initialization to runtime init")
Cc: Nabendu Maiti <nabendu.bikash.maiti@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk> (v2)
Cc: Ander Conselvan de Oliveira <conselvan2@gmail.com>
Cc: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Cc: Daniel Vetter <daniel.vetter@intel.com>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Cc: intel-gfx@lists.freedesktop.org
Signed-off-by: NAnder Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Tested-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: NVille Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1483365281-10569-1-git-send-email-ander.conselvan.de.oliveira@intel.com

Ville explained that the wakelock was being acquired during set-idle in
order to flush the voltage change from the punit.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170102152845.32352-1-chris@chris-wilson.co.ukReviewed-by: NVille Syrjälä <ville.syrjala@linux.intel.com>

This function is only used by intel_guc_send() and it doesn't
need to be exposed outside of intel_uc.o file. Also when defined
as static, compiler will generate smaller code. Additionally let
it take guc param instead dev_priv to match function name.
Signed-off-by: NMichal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Arkadiusz Hiler <arkadiusz.hiler@intel.com>
Reviewed-by: NTvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: NTvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161220115531.76120-1-michal.wajdeczko@intel.com

In future patches, we require greater flexibility in describing
the number of scalers available on each CRTC. To ease that transition
we move the current assignment to intel_device_info.

Scaler structure initialisation is done if scaler is available on the CRTC.
Gen9 check is not required as on depending upon numbers of scalers we
initialize scalers or return without doing anything in skl_init_scalers.

v3: Changed skl_init_scaler to intel_crtc_init_scalers

v2: Added Chris's comments.
Signed-off-by: NNabendu Maiti <nabendu.bikash.maiti@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v2)
Reviewed-by: NAnder Conselvan de Oliveira <conselvan2@gmail.com>
Signed-off-by: NAnder Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1480398794-22741-1-git-send-email-nabendu.bikash.maiti@intel.com

The function intel_atomic_get_shared_dpll_state() is only called from
intel_dpll_mgr.c and it concerns the same data structures as the other
functions in that file, so move it there and make it static.
Signed-off-by: NAnder Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Suggested-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1483024933-3726-8-git-send-email-ander.conselvan.de.oliveira@intel.com

Pull DAX updates from Dan Williams:
 "The completion of Jan's DAX work for 4.10.

  As I mentioned in the libnvdimm-for-4.10 pull request, these are some
  final fixes for the DAX dirty-cacheline-tracking invalidation work
  that was merged through the -mm, ext4, and xfs trees in -rc1. These
  patches were prepared prior to the merge window, but we waited for
  4.10-rc1 to have a stable merge base after all the prerequisites were
  merged.

  Quoting Jan on the overall changes in these patches:

     "So I'd like all these 6 patches to go for rc2. The first three
      patches fix invalidation of exceptional DAX entries (a bug which
      is there for a long time) - without these patches data loss can
      occur on power failure even though user called fsync(2). The other
      three patches change locking of DAX faults so that ->iomap_begin()
      is called in a more relaxed locking context and we are safe to
      start a transaction there for ext4"

  These have received a build success notification from the kbuild
  robot, and pass the latest libnvdimm unit tests. There have not been
  any -next releases since -rc1, so they have not appeared there"

* 'libnvdimm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm:
  ext4: Simplify DAX fault path
  dax: Call ->iomap_begin without entry lock during dax fault
  dax: Finish fault completely when loading holes
  dax: Avoid page invalidation races and unnecessary radix tree traversals
  mm: Invalidate DAX radix tree entries only if appropriate
  ext2: Return BH_New buffers for zeroed blocks

DRRS is not yet kerneldoc despite the allusion prior to enum
drrs_refresh_rate_type. Drop the '**' to avoid the warnings from
make htmldocs.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/20161231112012.29263-4-chris@chris-wilson.co.uk

The existing kerneldoc was outdated, so time for a refresh.

v2: Use single line kdoc, mention functions for manipulation
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/20161231112012.29263-3-chris@chris-wilson.co.uk

Parameter - no.
Parameter: yes.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/20161231112012.29263-2-chris@chris-wilson.co.uk

The read of the page pin count and the bind count are unordered,
presenting races in the assert and it firing off incorrectly. Prevent
this by restricting the assert to the vma bind/unbind routines where we
have local cpu ordering between the two.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/20161231112012.29263-1-chris@chris-wilson.co.uk

Pull documentation fixes from Jonathan Corbet:
 "Two small fixes:

   - A merge error on my part broke the DocBook build. I've
     requisitioned one of tglx's frozen sharks for appropriate
     disciplinary action and resolved to be more careful about testing
     the DocBook stuff as long as it's still around.

   - Fix an error in unaligned-memory-access.txt"

* tag 'docs-4.10-rc1-fix' of git://git.lwn.net/linux:
  Documentation/unaligned-memory-access.txt: fix incorrect comparison operator
  docs: Fix build failure

Pull crypto fix from Herbert Xu:
 "This fixes a boot failure on some platforms when crypto self test is
  enabled along with the new acomp interface"

* 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
  crypto: testmgr - Use heap buffer for acomp test input

Remove the IS_PLATFORM() macros from intel_dump_pipe_config() and split
that logic in platform specific implementations inside the dpll code,
accessed through a platform independent interface.

v2: Rebase.
Signed-off-by: NAnder Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch> (v1)
Link: http://patchwork.freedesktop.org/patch/msgid/1483024933-3726-7-git-send-email-ander.conselvan.de.oliveira@intel.com

The documentation for most of the non-static members and structs were
missing. Fix that.

v2: Fix typos (Durga)

v3: Rebase.
    Fix make docs warnings.
    Document more.

v4: capitilize CRTC; say that the prepare hook is a nop if the DPLL is
    already enabled; link to struct intel_dpll_hw_state from @hw_state
    field in struct intel_shared_dpll_state; reorganize DPLL flags; link
    intel_shared_dpll_state to other structs and functions. (Daniel)
Signed-off-by: NAnder Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Reviewed-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1483024933-3726-6-git-send-email-ander.conselvan.de.oliveira@intel.com

The hook is called from intel_prepare_shared_dpll(). The name doesn't
make sense after all the changes to modeset code. So just call it
prepare.
Signed-off-by: NAnder Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Reviewed-by: NDurgadoss R <durgadoss.r@intel.com>
Reviewed-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1483024933-3726-5-git-send-email-ander.conselvan.de.oliveira@intel.com

Struct intel_shared_dpll_config is used to hold the state of the DPLL in
the "atomic" sense, so call it state like everything else atomic.

v2: Rebase
Signed-off-by: NAnder Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch> (v1)
Link: http://patchwork.freedesktop.org/patch/msgid/1483024933-3726-4-git-send-email-ander.conselvan.de.oliveira@intel.com

The function intel_shared_dpll_commit() performs the equivalent of
drm_atomic_helper_swap_state() for the shared dpll state, which is not
handled by the helpers. So make it do a full swap of the state and
rename it for consistency.

v2: Fix typo in the commit message. (Durga)
v3: Rebase.
v4: Swap the states instead of just renaming the function. (Daniel)
Signed-off-by: NAnder Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Reviewed-by: Durgadoss R <durgadoss.r@intel.com> (v2)
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch> (v3)
Link: http://patchwork.freedesktop.org/patch/msgid/1483024933-3726-3-git-send-email-ander.conselvan.de.oliveira@intel.com

While the details of getting a shared dpll are wrapped by
intel_get_shared_dpll(), the release was still hand rolled into the
modeset code. Fix that by creating an entry point for releasing the
pll and move that code there.

v2: Take old_dpll from crtc->state instead of crtc_state. (CI)
Signed-off-by: NAnder Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Reviewed-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1483024933-3726-2-git-send-email-ander.conselvan.de.oliveira@intel.com

 mm/filemap.c: In function 'clear_bit_unlock_is_negative_byte':
  mm/filemap.c:933:9: error: too few arguments to function 'test_bit'
    return test_bit(PG_waiters);
         ^~~~~~~~

Fixes: b91e1302 ('mm: optimize PageWaiters bit use for unlock_page()')
Signed-off-by: NOlof Johansson <olof@lixom.net>
Brown-paper-bag-by: NLinus Torvalds <dummy@duh.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

In commit 62906027 ("mm: add PageWaiters indicating tasks are
waiting for a page bit") Nick Piggin made our page locking no longer
unconditionally touch the hashed page waitqueue, which not only helps
performance in general, but is particularly helpful on NUMA machines
where the hashed wait queues can bounce around a lot.

However, the "clear lock bit atomically and then test the waiters bit"
sequence turns out to be much more expensive than it needs to be,
because you get a nasty stall when trying to access the same word that
just got updated atomically.

On architectures where locking is done with LL/SC, this would be trivial
to fix with a new primitive that clears one bit and tests another
atomically, but that ends up not working on x86, where the only atomic
operations that return the result end up being cmpxchg and xadd.  The
atomic bit operations return the old value of the same bit we changed,
not the value of an unrelated bit.

On x86, we could put the lock bit in the high bit of the byte, and use
"xadd" with that bit (where the overflow ends up not touching other
bits), and look at the other bits of the result.  However, an even
simpler model is to just use a regular atomic "and" to clear the lock
bit, and then the sign bit in eflags will indicate the resulting state
of the unrelated bit #7.

So by moving the PageWaiters bit up to bit #7, we can atomically clear
the lock bit and test the waiters bit on x86 too.  And architectures
with LL/SC (which is all the usual RISC suspects), the particular bit
doesn't matter, so they are fine with this approach too.

This avoids the extra access to the same atomic word, and thus avoids
the costly stall at page unlock time.

The only downside is that the interface ends up being a bit odd and
specialized: clear a bit in a byte, and test the sign bit.  Nick doesn't
love the resulting name of the new primitive, but I'd rather make the
name be descriptive and very clear about the limitation imposed by
trying to work across all relevant architectures than make it be some
generic thing that doesn't make the odd semantics explicit.

So this introduces the new architecture primitive

    clear_bit_unlock_is_negative_byte();

and adds the trivial implementation for x86.  We have a generic
non-optimized fallback (that just does a "clear_bit()"+"test_bit(7)"
combination) which can be overridden by any architecture that can do
better.  According to Nick, Power has the same hickup x86 has, for
example, but some other architectures may not even care.

All these optimizations mean that my page locking stress-test (which is
just executing a lot of small short-lived shell scripts: "make test" in
the git source tree) no longer makes our page locking look horribly bad.
Before all these optimizations, just the unlock_page() costs were just
over 3% of all CPU overhead on "make test".  After this, it's down to
0.66%, so just a quarter of the cost it used to be.

(The difference on NUMA is bigger, but there this micro-optimization is
likely less noticeable, since the big issue on NUMA was not the accesses
to 'struct page', but the waitqueue accesses that were already removed
by Nick's earlier commit).
Acked-by: NNick Piggin <npiggin@gmail.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Bob Peterson <rpeterso@redhat.com>
Cc: Steven Whitehouse <swhiteho@redhat.com>
Cc: Andrew Lutomirski <luto@kernel.org>
Cc: Andreas Gruenbacher <agruenba@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>