drm/i915: Use remap_io_mapping() to prefault all PTE in a single pass

Very old numbers indicate this is a 66% improvement when remapping the
entire object for fence contention - due to the elimination of
track_pfn_insert and its strcmp.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Testcase: igt/gem_fence_upload/performance
Testcase: igt/gem_mmap_gtt
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160819155428.1670-6-chris@chris-wilson.co.uk

drm/i915: Embed the io-mapping struct inside drm_i915_private

As io_mapping.h now always allocates the struct, we can avoid that
allocation and extra pointer dance by embedding the struct inside
drm_i915_private

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160819155428.1670-5-chris@chris-wilson.co.uk

io-mapping: Always create a struct to hold metadata about the io-mapping

Currently, we only allocate a structure to hold metadata if we need to
allocate an ioremap for every access, such as on x86-32. However, it
would be useful to store basic information about the io-mapping, such as
its page protection, on all platforms.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: linux-mm@kvack.org
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160819155428.1670-4-chris@chris-wilson.co.uk

drm/i915/fbc: Allow on unfenced surfaces, for recent gen

Only fbc1 is tied to using a fence. Later iterations of fbc are more
flexible and allow operation on unfenced frontbuffers.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@intel.com>
Cc: "Zanoni, Paulo R" <paulo.r.zanoni@intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160819155428.1670-3-chris@chris-wilson.co.uk

drm/i915/fbc: Don't set an illegal fence if unfenced

If the frontbuffer doesn't have an associated fence, it will have a
fence reg of -1. If we attempt to OR in this register into the FBC
control register we end up setting all control bits, oops!

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Daniel Vetter <daniel.vetter@intel.com>
Cc: "Zanoni, Paulo R" <paulo.r.zanoni@intel.com>
Reviwed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160819155428.1670-2-chris@chris-wilson.co.uk

drm/i915: Flush delayed fence releases after reset

What I never hit in testing, but Mika immediately did, was a GPU hang
with a pending fence release (where a tiled object has been changed by
the user to be untiled, and the update has not yet been committed to the
fence register). As the stride/tiling is 0, this causes a divide-by-zero
error when trying to write the new fence parameters:

[   28.784518] drm/i915: Resetting chip after gpu hang
[   28.784551] divide error: 0000 [#1] PREEMPT SMP
[   28.784565] Modules linked in: nls_iso8859_1 nls_cp437 vfat fat mxm_wmi x86_pkg_temp_thermal snd_hda_codec_hdmi kvm irqbypass snd_hda_codec_realtek snd_hda_codec_generic snd_hda_intel snd_hda_codec serio_raw snd_hwdep snd_hda_core snd_pcm snd_seq_midi snd_seq_midi_event snd_rawmidi snd_seq snd_timer snd_seq_device snd soundcore mac_hid wmi efivarfs autofs4 raid10 raid456 libcrc32c async_raid6_recov async_memcpy async_pq raid6_pq async_xor xor async_tx raid0 multipath linear psmouse e1000e ptp pps_core nvme nvme_core i915 i2c_algo_bit drm_kms_helper syscopyarea sysfillrect sysimgblt fb_sys_fops drm video
[   28.784738] CPU: 0 PID: 1692 Comm: kworker/0:2 Not tainted 4.8.0-rc2+ #895
[   28.784752] Hardware name: System manufacturer System Product Name/Z170M-PLUS, BIOS 1803 05/09/2016
[   28.784786] Workqueue: events_long i915_hangcheck_elapsed [i915]
[   28.784814] task: ffff923c18f59d40 task.stack: ffff923c1b7e4000
[   28.784827] RIP: 0010:[<ffffffffc0475b5f>]  [<ffffffffc0475b5f>] fence_write+0x9f/0x3b0 [i915]
[   28.784854] RSP: 0018:ffff923c1b7e7b30  EFLAGS: 00010246
[   28.784866] RAX: 00000000008ca000 RBX: ffff923c18540000 RCX: 0000000000000020
[   28.784880] RDX: 0000000000000000 RSI: 0000000000000000 RDI: 000000000596d000
[   28.784894] RBP: ffff923c1b7e7b68 R08: 0000000000000000 R09: 0000000000000000
[   28.784908] R10: 0000000000000000 R11: 00000000008ca000 R12: ffff923c1ef9d600
[   28.784921] R13: 0000000000100040 R14: 0000000000100044 R15: ffff923c18549908
[   28.784935] FS:  0000000000000000(0000) GS:ffff923c36c00000(0000) knlGS:0000000000000000
[   28.784951] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   28.784962] CR2: 00007f193373c893 CR3: 0000000419c78000 CR4: 00000000003406f0
[   28.784976] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[   28.784990] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[   28.785004] Stack:
[   28.785009]  000000000596c03b ffff923c1b7e7b68 ffff923c18549938 0000000000000009
[   28.785026]  ffff923c18540000 ffff923c18549280 ffff923c18547ce8 ffff923c1b7e7b90
[   28.785044]  ffffffffc04761f9 ffff923c18540000 ffff923c18547d00 ffff923c18548ff8
[   28.785062] Call Trace:
[   28.785078]  [<ffffffffc04761f9>] i915_gem_restore_fences+0x39/0x50 [i915]
[   28.785102]  [<ffffffffc047fe89>] i915_gem_reset+0x179/0x300 [i915]

Reported-by: Mika Kuoppala <mika.kuoppala@intel.com>
Fixes: 49ef5294cda2 ("drm/i915: Move fence tracking from object to vma")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160819155428.1670-1-chris@chris-wilson.co.uk

drm/i915: Reattach comment, complete type specification

In the recent patch
bc3d674 drm/i915: Allow userspace to request no-error-capture upon ...
the final version moved the flags and the associated #defines around
so they were adjacent; unfortunately, they ended up between a comment
and the thing (hw_id) to which the comment applies :(

So this patch reshuffles the comment and subject back together.

Also, as we're touching 'hw_id', let's change it from just 'unsigned'
to a fully-specified 'unsigned int', because some code checking tools
(including checkpatch) object to plain 'unsigned'.

Fixes: bc3d674462e5 ("drm/i915: Allow userspace to request no-error-capture...")
Signed-off-by: Dave Gordon <david.s.gordon@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/1471616622-6919-1-git-send-email-david.s.gordon@intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915/cmdparser: Accelerate copies from WC memory

If we need to use clflush to prepare our batch for reads from memory, we
can bypass the cache instead by using non-temporal copies.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.william.auld@gmail.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-39-chris@chris-wilson.co.uk

drm/i915/cmdparser: Use binary search for faster register lookup

A significant proportion of the cmdparsing time for some batches is the
cost to find the register in the mmiotable. We ensure that those tables
are in ascending order such that we could do a binary search if it was
ever merited. It is.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-38-chris@chris-wilson.co.uk

drm/i915/cmdparser: Check for SKIP descriptors first

If the command descriptor says to skip it, ignore checking for anyother
other conflict.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-37-chris@chris-wilson.co.uk

drm/i915/cmdparser: Compare against the previous command descriptor

On the blitter (and in test code), we see long sequences of repeated
commands, e.g. XY_PIXEL_BLT, XY_SCANLINE_BLT, or XY_SRC_COPY. For these,
we can skip the hashtable lookup by remembering the previous command
descriptor and doing a straightforward compare of the command header.
The corollary is that we need to do one extra comparison before lookup
up new commands.

v2: Less magic mask (ok, it is still magic, but now you cannot see!)

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-36-chris@chris-wilson.co.uk

drm/i915/cmdparser: Improve hash function

The existing code's hashfunction is very suboptimal (most 3D commands
use the same bucket degrading the hash to a long list). The code even
acknowledge that the issue was known and the fix simple:

/*
 * If we attempt to generate a perfect hash, we should be able to look at bits
 * 31:29 of a command from a batch buffer and use the full mask for that
 * client. The existing INSTR_CLIENT_MASK/SHIFT defines can be used for this.
 */

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-35-chris@chris-wilson.co.uk

drm/i915/cmdparser: Only cache the dst vmap

For simplicity, we want to continue using a contiguous mapping of the
command buffer, but we can reduce the number of vmappings we hold by
switching over to a page-by-page copy from the user batch buffer to the
shadow. The cost for saving one linear mapping is about 5% in trivial
workloads - which is more or less the overhead in calling kmap_atomic().

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-34-chris@chris-wilson.co.uk

drm/i915/cmdparser: Use cached vmappings

The single largest factor in the overhead of parsing the commands is the
setup of the virtual mapping to provide a continuous block for the batch
buffer. If we keep those vmappings around (against the better judgement
of mm/vmalloc.c, which we offset by handwaving and looking suggestively
at the shrinker) we can dramatically improve the performance of the
parser for small batches (such as media workloads). Furthermore, we can
use the prepare shmem read/write functions to determine  how best we
need to clflush the range (rather than every page of the object).

The impact of caching both src/dst vmaps is +80% on ivb and +140% on byt
for the throughput on small batches. (Caching just the dst vmap and
iterating over the src, doing a page by page copy is roughly 5% slower
on both platforms. That may be an acceptable trade-off to eliminate one
cached vmapping, and we may be able to reduce the per-page copying overhead
further.) For *this* simple test case, the cmdparser is now within a
factor of 2 of ideal performance.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.william.auld@gmail.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-33-chris@chris-wilson.co.uk

drm/i915/cmdparser: Add the TIMESTAMP register for the other engines

Since I have been using the BCS_TIMESTAMP to measure latency of
execution upon the blitter ring, allow regular userspace to also read
from that register. They are already allowed RCS_TIMESTAMP!

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-32-chris@chris-wilson.co.uk

drm/i915/cmdparser: Make initialisation failure non-fatal

If the developer adds a register in the wrong order, we BUG during boot.
That makes development and testing very difficult. Let's be a bit more
friendly and disable the command parser with a big warning if the tables
are invalid.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-31-chris@chris-wilson.co.uk

drm/i915: Stop discarding GTT cache-domain on unbind vma

Since commit 43566dedde54 ("drm/i915: Broaden application of
set-domain(GTT)") we allowed objects to be in the GTT domain, but unbound.
Therefore removing the GTT cache domain when removing the GGTT vma is no
longer semantically correct.

An unfortunate side-effect is we lose the wondrously named
i915_gem_object_finish_gtt(), not to be confused with
i915_gem_gtt_finish_object()!

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Akash Goel <akash.goel@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-30-chris@chris-wilson.co.uk

drm/i915: Bump the inactive tracking for all VMA accessed

We track the LRU access for eviction and bump the last access for the
user GGTT on set-to-gtt. When we do so we need to not only bump the
primary GGTT VMA but all partials as well. Similarly we want to
bump the last access tracking for when unpinning an object from the
scanout so that they do not get promptly evicted and hopefully remain
available for reuse on the next frame.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-29-chris@chris-wilson.co.uk

drm/i915: Track display alignment on VMA

When using the aliasing ppgtt and pageflipping with the shrinker/eviction
active, we note that we often have to rebind the backbuffer before
flipping onto the scanout because it has an invalid alignment. If we
store the worst-case alignment required for a VMA, we can avoid having
to rebind at critical junctures.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-28-chris@chris-wilson.co.uk

drm/i915: Fallback to using unmappable memory for scanout

The existing ABI says that scanouts are pinned into the mappable region
so that legacy clients (e.g. old Xorg or plymouthd) can write directly
into the scanout through a GTT mapping. However if the surface does not
fit into the mappable region, we are better off just trying to fit it
anywhere and hoping for the best. (Any userspace that is capable of
using ginormous scanouts is also likely not to rely on pure GTT
updates.) With the partial vma fault support, we are no longer
restricted to only using scanouts that we can pin (though it is still
preferred for performance reasons and for powersaving features like
FBC).

v2: Skip fence pinning when not mappable.
v3: Add a comment to explain the possible ramifications of not being
    able to use fences for unmappable scanouts.
v4: Rebase to skip over some local patches
v5: Rebase to defer until after we have unmappable GTT fault support

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Deepak S <deepak.s@linux.intel.com>
Cc: Damien Lespiau <damien.lespiau@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-27-chris@chris-wilson.co.uk

drm/i915: Choose not to evict faultable objects from the GGTT

Often times we do not want to evict mapped objects from the GGTT as
these are quite expensive to teardown and frequently reused (causing an
equally, if not more so, expensive setup). In particular, when faulting
in a new object we want to avoid evicting an active object, or else we
may trigger a page-fault-of-doom as we ping-pong between evicting two
objects.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-26-chris@chris-wilson.co.uk

drm/i915: Drop ORIGIN_GTT for untracked GTT writes

If FBC is set on a framebuffer that is unmapped, all GTT faults will be
from a partial mapping. Writes by the user through the partial VMA are
then untracked by the FBC and so we must use the ORIGIN_CPU when flushing
the I915_GEM_DOMAIN_GTT.

v2: Keep ORIGIN_CPU for set-to-domain(.write=CPU)

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Daniel Vetter <daniel.vetter@intel.com>
Cc: "Zanoni, Paulo R" <paulo.r.zanoni@intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-25-chris@chris-wilson.co.uk

drm/i915: Convert partial ggtt vma to full ggtt if it spans the entire object

If we want to create a partial vma from a chunk that is the same size as
the object, create a normal ggtt vma instead. The benefit is that it
will match future requests for the normal ggtt.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-24-chris@chris-wilson.co.uk

drm/i915: Fix partial GGTT faulting

We want to always use the partial VMA as a fallback for a failure to
bind the object into the GGTT. This extends the support partial objects
in the GGTT to cover everything, not just objects too large.

v2: Call the partial view, view not partial.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-23-chris@chris-wilson.co.uk

drm/i915: Choose partial chunksize based on tile row size

In order to support setting up fences for partial mappings of an object,
we have to align those mappings with the fence. The minimum chunksize we
choose is at least the size of a single tile row.

v2: Make minimum chunk size a define for later use

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-22-chris@chris-wilson.co.uk

drm/i915: Move fence tracking from object to vma

In order to handle tiled partial GTT mmappings, we need to associate the
fence with an individual vma.

v2: A couple of silly drops replaced spotted by Joonas

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-21-chris@chris-wilson.co.uk

drm/i915: Rename fence.lru_list to link

Our current practice is to only name the actual list (here
dev_priv->fence_list) using "list", and elements upon that list are
referred to as "link". Further, the lru nature is of the list and not of
the node and including in the name does not disambiguate the link from
anything else.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-20-chris@chris-wilson.co.uk

drm/i915/userptr: Make gup errors stickier

Keep any error reported by the gup_worker until we are notified that the
arena has changed (via the mmu-notifier). This has the importance of
making two consecutive calls to i915_gem_object_get_pages() reporting
the same error, and curtailing a loop of detecting a fault and requeueing
a gup_worker.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-19-chris@chris-wilson.co.uk

drm/i915: Allocate rings from stolen

If we have stolen available, make use of it for ringbuffer allocation.
Previously this was restricted to !llc platforms, as writing to stolen
requires a GGTT mapping - but now that we have partial mappable support,
the mappable aperture isn't quite so precious so we can use it more
freely and ringbuffers are a good user for the otherwise wasted stolen.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-18-chris@chris-wilson.co.uk

drm/i915: Allow ringbuffers to be bound anywhere

Now that we have WC vmapping available, we can bind our rings anywhere
in the GGTT and do not need to restrict them to the mappable region.
Except for stolen objects, for which direct access is verbatim and we
must use the mappable aperture.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-17-chris@chris-wilson.co.uk

drm/i915: Move map-and-fenceable tracking to the VMA

By moving map-and-fenceable tracking from the object to the VMA, we gain
fine-grained tracking and the ability to track individual fences on the VMA
(subsequent patch).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-16-chris@chris-wilson.co.uk

drm/i915: Disallow direct CPU access to stolen pages for relocations

As we cannot access the backing pages behind stolen objects, we should
not attempt to do so for relocations.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-15-chris@chris-wilson.co.uk

drm/i915: Fallback to single page GTT mmappings for relocations

If we cannot pin the entire object into the mappable region of the GTT,
try to pin a single page instead. This is much more likely to succeed,
and prevents us falling back to the clflush slow path.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-14-chris@chris-wilson.co.uk

drm/i915: Refactor execbuffer relocation writing

With the introduction of the reloc page cache, we are just one step away
from refactoring the relocation write functions into one. Not only does
it tidy the code (slightly), but it greatly simplifies the control logic
much to gcc's satisfaction.

v2: Add selftests to document the relationship between the clflush
flags, the KMAP bit and packing into the page-aligned pointer.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-13-chris@chris-wilson.co.uk

drm/i915: Tidy up flush cpu/gtt write domains

Since we know the write domain, we can drop the local variable and make
the code look a tiny bit simpler.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-12-chris@chris-wilson.co.uk

drm/i915: Pin the pages first in shmem prepare read/write

There is an improbable, but not impossible, case that if we leave the
pages unpin as we operate on the object, then somebody via the shrinker
may steal the lock (which lock? right now, it is struct_mutex, THE lock)
and change the cache domains after we have already inspected them.

(Whilst here, avail ourselves of the opportunity to take a couple of
steps to make the two functions look more similar.)

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-11-chris@chris-wilson.co.uk

drm/i915: Wait for writes through the GTT to land before reading back

If we quickly switch from writing through the GTT to a read of the
physical page directly with the CPU (e.g. performing relocations through
the GTT and then running the command parser), we can observe that the
writes are not visible to the CPU. It is not a coherency problem, as
extensive investigations with clflush have demonstrated, but a mere
timing issue - we have to wait for the GTT to complete it's write before
we start our read from the CPU.

The issue can be illustrated in userspace with:

gtt = gem_mmap__gtt(fd, handle, 0, OBJECT_SIZE, PROT_READ | PROT_WRITE);
cpu = gem_mmap__cpu(fd, handle, 0, OBJECT_SIZE, PROT_READ | PROT_WRITE);
gem_set_domain(fd, handle, I915_GEM_DOMAIN_GTT, I915_GEM_DOMAIN_GTT);

for (i = 0; i < OBJECT_SIZE / 64; i++) {
int x = 16*i + (i%16);
gtt[x] = i;
clflush(&cpu[x], sizeof(cpu[x]));
assert(cpu[x] == i);
}

Experimenting with that shows that this behaviour is indeed limited to
recent Atom-class hardware.

Testcase: igt/gem_exec_flush/basic-batch-default-cmd #byt
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-10-chris@chris-wilson.co.uk

drm/i915: Before accessing an object via the cpu, flush GTT writes

If we want to read the pages directly via the CPU, we have to be sure
that we have to flush the writes via the GTT (as the CPU can not see
the address aliasing).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-9-chris@chris-wilson.co.uk

drm/i915: Extract i915_gem_obj_prepare_shmem_write()

This is a companion to i915_gem_obj_prepare_shmem_read() that prepares
the backing storage for direct writes. It first serialises with the GPU,
pins the backing storage and then indicates what clfushes are required in
order for the writes to be coherent.

Whilst here, fix support for ancient CPUs without clflush for which we
cannot do the GTT+clflush tricks.

v2: Add i915_gem_obj_finish_shmem_access() for symmetry

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-8-chris@chris-wilson.co.uk

drm/i915: Cache kmap between relocations

When doing relocations, we have to obtain a mapping to the page
containing the target address. This is either a kmap or iomap depending
on GPU and its cache coherency. Neighbouring relocation entries are
typically within the same page and so we can cache our kmapping between
them and avoid those pesky TLB flushes.

Note that there is some sleight-of-hand in how the slow relocate works
as the reloc_entry_cache implies pagefaults disabled (as we are inside a
kmap_atomic section). However, the slow relocate code is meant to be the
fallback from the atomic fast path failing. Fortunately it works as we
already have performed the copy_from_user for the relocation array (no
more pagefaults there) and the kmap_atomic cache is enabled after we
have waited upon an active buffer (so no more sleeping in atomic).
Magic!

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-7-chris@chris-wilson.co.uk

drm/i915: Fallback to single page pwrite/pread if unable to release fence

If we cannot release the fence (for example if someone is inexplicably
trying to write into a tiled framebuffer that is currently pinned to the
display! *cough* kms_frontbuffer_tracking *cough*) fallback to using the
page-by-page pwrite/pread interface, rather than fail the syscall
entirely.

Since this is triggerable by the user (along pwrite) we have to remove
the WARN_ON(fence->pin_count).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-6-chris@chris-wilson.co.uk

drm/i915: Mark up the GTT flush following WC writes as ORIGIN_CPU

Similarly to invalidating beforehand, if the object is mmapped via
I915_MMAP_WC we cannot track writes through the I915_GEM_DOMAIN_GTT. At
the conclusion of the write, i915_gem_object_flush_gtt_writes() we also
need to treat the origin carefully in case it may have been untracked.

See also commit aeecc9696aa0 ("drm/i915: use ORIGIN_CPU for frontbuffer
invalidation on WC mmaps").

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-5-chris@chris-wilson.co.uk

drm/i915: Use ORIGIN_CPU for fb invalidation from pwrite

As pwrite does not use the fence for its GTT access, and may even go
through a secondary interface avoiding the main VMA, we cannot treat the
write as automatically invalidated by the hardware and so we require
ORIGIN_CPU frontbufer invalidate/flushes.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-4-chris@chris-wilson.co.uk
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: vfree() no longer ignores the low bits of the address

Since vfree() now likes to WARN when passed a non-page-aligned pointer,
we need to discard the low bits to comply with it.

Fixes: d31d7cb1460c ("drm/i915: Support for creating write combined type vmaps")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-3-chris@chris-wilson.co.uk

agp/intel: Flush chipset writes after updating a single PTE

After we update one PTE for a page, the caller expects to be able to
immediately use that through a GGTT read/write. To comply with the
callers expectations we therefore need to flush the chipset buffers
before returning.

Reported-by: Matti Hämäläinen <ccr@tnsp.org>
Fixes: d6473f566417 ("drm/i915: Add support for mapping an object page...")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ankitprasad Sharma <ankitprasad.r.sharma@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Tested-by: Matti Hämäläinen <ccr@tnsp.org>
Cc: drm-intel-fixes@lists.freedesktop.org
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-2-chris@chris-wilson.co.uk

drm/i915: Unconditionally flush any chipset buffers before execbuf

If userspace is asynchronously streaming into the batch or other
execobjects, we may not flush those writes along with a change in cache
domain (as there is no change). Therefore those writes may end up in
internal chipset buffers and not visible to the GPU upon execution. We
must issue a flush command or otherwise we encounter incoherency in the
batchbuffers and the GPU executing invalid commands (i.e. hanging) quite
regularly.

v2: Throw a paranoid wmb() into the general flush so that we remain
consistent with before.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=90841
Fixes: 1816f9236303 ("drm/i915: Support creation of unbound wc user...")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Akash Goel <akash.goel@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Tested-by: Matti Hämäläinen <ccr@tnsp.org>
Cc: stable@vger.kernel.org
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160818161718.27187-1-chris@chris-wilson.co.uk

drm/i915/gen9: Drop invalid WARN() during data rate calculation

It's possible to have a non-zero plane mask and still wind up with a
total data rate of zero.  There are two cases where this can happen:

 * planes are active (from the KMS point of view), but are
   all fully clipped (positioned offscreen)
 * the only active plane on a CRTC is the cursor (which is handled
   independently and not counted into the general data rate computations

These are both valid display setups (although unusual), so we need to
drop the WARN().

Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Testcase: kms_universal_planes.cursor-only-pipe-*
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1466196140-16336-4-git-send-email-matthew.d.roper@intel.com
Cc: stable@vger.kernel.org #v4.7+

drm/i915/gen9: Initialize intel_state->active_crtcs during WM sanitization (v2)

intel_state->active_crtcs is usually only initialized when doing a
modeset.  During our first atomic commit after boot, we're effectively
faking a modeset to sanitize the DDB/wm setup, so ensure that this field
gets initialized before use.

v2:
 - Don't clobber active_crtcs if our first commit really is a modeset
   (Maarten)
 - Grab connection_mutex when faking a modeset during sanitization
   (Maarten)

Reported-by: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Tested-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1466196140-16336-2-git-send-email-matthew.d.roper@intel.com
Cc: stable@vger.kernel.org #v4.7+

drm/i915: Add missing kerneldoc for guc_client_alloc:engines

Brief parameter text to describe @engines.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Dave Gordon <david.s.gordon@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471437762-22936-1-git-send-email-chris@chris-wilson.co.uk
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

drm/i915: Remember to set vma->pages for the preallocated stolen object

In commit 247177ddd517 ("drm/i915: Always set the vma->pages"), as it
title implies, we always set vma->pages for bound objects. Even before
that, we would set vma->ggtt_view.pages, for globally bound objects.
This was forgotten for the fixup inside the preallocated stolen objects,
which has to recreate a global GTT binding outside of the usual VMA
insertion path

Fixes: 247177ddd517 ("drm/i915: Always set the vma->pages")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471430013-3449-1-git-send-email-chris@chris-wilson.co.uk
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

drm/i915: Mark i915_hpd_poll_init_work as static

Local function with forgotten static declaration.

Fixes: 19625e85c6ec ("drm/i915: Enable polling when we don't have hpd")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Lyude <cpaul@redhat.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471432146-5196-2-git-send-email-chris@chris-wilson.co.uk
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

drm/i915: Mark the static key for movntqda as static

Silence sparse who warns that the global variable is not declared
static.

Fixes: 0b1de5d58e19 ("drm/i915: Use SSE4.1 movntdqa to ...")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Akash Goel <akash.goel@intel.com>
Cc: Damien Lespiau <damien.lespiau@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471432146-5196-1-git-send-email-chris@chris-wilson.co.uk
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

drm/i915: Initialize legacy semaphores from engine hw id indexed array

Build the legacy semaphore initialisation array using the engine
hardware ids instead of driver internal ones. This makes the
static array size dependent only on the number of gen6 semaphore
engines.

Also makes the per-engine semaphore wait and signal tables
hardware id indexed saving some more space.

v2: Refactor I915_GEN6_NUM_ENGINES to GEN6_SEMAPHORE_LAST. (Chris Wilson)
v3: More polish. (Chris Wilson)

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/1471363461-9973-1-git-send-email-tvrtko.ursulin@linux.intel.com

drm/i915: Add enum for hardware engine identifiers

Put the engine hardware id in the common header so they are
not only associated with the GuC since they are needed for
the legacy semaphores implementation.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Silence GCC warning for cmn_a_well

Just make the logic simple enough for even GCC to understand (and
foolproof against random changes):

drivers/gpu/drm/i915/intel_runtime_pm.c: warning: 'cmn_a_well' may be
used uninitialized in this function [-Wuninitialized]:  => 871:23

Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/1471284383-22324-1-git-send-email-chris@chris-wilson.co.uk
Reviewed-by: Imre Deak <imre.deak@intel.com>

drm/i915: Embrace the race in busy-ioctl

Daniel Vetter proposed a new challenge to the serialisation inside the
busy-ioctl that exposed a flaw that could result in us reporting the
wrong engine as being busy. If the request is reallocated as we test
its busyness and then reassigned to this object by another thread, we
would not notice that the test itself was incorrect.

We are faced with a choice of using __i915_gem_active_get_request_rcu()
to first acquire a reference to the request preventing the race, or to
acknowledge the race and accept the limitations upon the accuracy of the
busy flags. Note that we guarantee that we never falsely report the
object as idle (providing userspace itself doesn't race), and so the
most important use of the busy-ioctl and its guarantees are fulfilled.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Daniel Vetter <daniel.vetter@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471337440-16777-1-git-send-email-chris@chris-wilson.co.uk

drm/i915: Initialise mmaped_count for i915_gem_object_info

Reported-by: 0day kbuild test robot
Fixes: 2bd160a131ac ("drm/i915: Reduce i915_gem_objects to only show...")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/1471263496-27537-1-git-send-email-chris@chris-wilson.co.uk
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>

drm/i915: Record the RING_MODE register for post-mortem debugging

Just another useful register to inspect following a GPU hang.

v2: Remove partial decoding of RING_MODE to userspace, be consistent and
use GEN > 2 guards around RING_MODE everywhere.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-32-git-send-email-chris@chris-wilson.co.uk

drm/i915: Only record active and pending requests upon a GPU hang

There is no other state pertaining to the completed requests in the
hang, other than gleamed through the ringbuffer, so including the
expired requests in the list of outstanding requests simply adds noise.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-31-git-send-email-chris@chris-wilson.co.uk

drm/i915: Print the batchbuffer offset next to BBADDR in error state

It is useful when looking at captured error states to check the recorded
BBADDR register (the address of the last batchbuffer instruction loaded)
against the expected offset of the batch buffer, and so do a quick check
that (a) the capture is true or (b) HEAD hasn't wandered off into the
badlands.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-30-git-send-email-chris@chris-wilson.co.uk

drm/i915: Move debug only per-request pid tracking from request to ctx

Since contexts are not currently shared between userspace processes, we
have an exact correspondence between context creator and guilty batch
submitter. Therefore we can save some per-batch work by inspecting the
context->pid upon error instead. Note that we take the context's
creator's pid rather than the file's pid in order to better track fd
passed over sockets.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-29-git-send-email-chris@chris-wilson.co.uk

drm/i915: Introduce i915_ggtt_offset()

This little helper only exists to safely discard the upper unused 32bits
of the general 64-bit VMA address - as we know that all Global GTT
currently are less than 4GiB in size and so that the upper bits must be
zero. In many places, we use a u32 for the global GTT offset and we want
to document where we are discarding the full VMA offset.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-28-git-send-email-chris@chris-wilson.co.uk

drm/i915: Track pinned VMA

Treat the VMA as the primary struct responsible for tracking bindings
into the GPU's VM. That is we want to treat the VMA returned after we
pin an object into the VM as the cookie we hold and eventually release
when unpinning. Doing so eliminates the ambiguity in pinning the object
and then searching for the relevant pin later.

v2: Joonas' stylistic nitpicks, a fun rebase.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-27-git-send-email-chris@chris-wilson.co.uk

drm/i915: Consolidate i915_vma_unpin_and_release()

In a few places, we repeat a call to clear a pointer to a vma whilst
unpinning and releasing a reference to its owner. Refactor those into a
common function.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-26-git-send-email-chris@chris-wilson.co.uk

drm/i915: Use VMA for wa_ctx tracking

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-25-git-send-email-chris@chris-wilson.co.uk

drm/i915: Use VMA for render state page tracking

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-24-git-send-email-chris@chris-wilson.co.uk

drm/i915: Use VMA as the primary tracker for semaphore page

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-23-git-send-email-chris@chris-wilson.co.uk

drm/i915/overlay: Use VMA as the primary tracker for images

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-22-git-send-email-chris@chris-wilson.co.uk

drm/i915: Move common seqno reset to intel_engine_cs.c

Since the intel_engine_init_seqno() is shared by all engine submission
backends, move it out of the legacy intel_ringbuffer.c and
into the new home for common routines, intel_engine_cs.c

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-21-git-send-email-chris@chris-wilson.co.uk

drm/i915: Move common scratch allocation/destroy to intel_engine_cs.c

Since the scratch allocation and cleanup is shared by all engine
submission backends, move it out of the legacy intel_ringbuffer.c and
into the new home for common routines, intel_engine_cs.c

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-20-git-send-email-chris@chris-wilson.co.uk

drm/i915: Use VMA for scratch page tracking

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-19-git-send-email-chris@chris-wilson.co.uk

drm/i915: Use VMA for ringbuffer tracking

Use the GGTT VMA as the primary cookie for handing ring objects as
the most common action upon the ring is mapping and unmapping which act
upon the VMA itself. By restructuring the code to work with the ring
VMA, we can shrink the code and remove a few cycles from context pinning.

v2: Move the flush of the object back to before the first pin. We use
the am-I-bound? query to only have to check the flush on the first
bind and so avoid stalling on active rings.
Lots of little renames and small hoops.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-18-git-send-email-chris@chris-wilson.co.uk

drm/i915: Move assertion for iomap access to i915_vma_pin_iomap

Access through the GTT requires the device to be awake. Ideally
i915_vma_pin_iomap() is short-lived and the pinning demarcates the
access through the iomap. This is not entirely true, we have a mixture
of long lived pins that exceed the wakelock (such as legacy ringbuffers)
and short lived pin that do live within the wakelock (such as execlist
ringbuffers).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-17-git-send-email-chris@chris-wilson.co.uk

drm/i915: Only change the context object's domain when binding

We know that the only access to the context object is via the GPU, and
the only time when it can be out of the GPU domain is when it is swapped
out and unbound. Therefore we only need to clflush the object when
binding, thus avoiding any potential stall on touching the domain on an
active context.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-16-git-send-email-chris@chris-wilson.co.uk

drm/i915: Use VMA as the primary object for context state

When working with contexts, we most frequently want the GGTT VMA for the
context state, first and foremost. Since the object is available via the
VMA, we need only then store the VMA.

v2: Formatting tweaks to debugfs output, restored some comments removed
in the next patch

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-15-git-send-email-chris@chris-wilson.co.uk

drm/i915: Use VMA directly for checking tiling parameters

v2: Rename functions to suit their more active role

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-14-git-send-email-chris@chris-wilson.co.uk

drm/i915: Convert fence computations to use vma directly

Lookup the GGTT vma once for the object assigned to the fence, and then
derive everything from that vma.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-13-git-send-email-chris@chris-wilson.co.uk

drm/i915: Track pinned vma inside guc

Since the guc allocates and pins and object into the GGTT for its usage,
it is more natural to use that pinned VMA as our resource cookie.

v2: Embrace naming tautology
v3: Rewrite comments for guc_allocate_vma()

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-12-git-send-email-chris@chris-wilson.co.uk

drm/i915: Add convenience wrappers for vma's object get/put

The VMA are unreferenced, they belong to the object and live until they
are closed. However, if we want to use the VMA as a cookie and use it to
keep the object alive, we want to hold onto a reference to the object
for the lifetime of the VMA cookie. To facilitate this, add a couple of
simple wrappers for managing the reference count on the object owning the
VMA.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-11-git-send-email-chris@chris-wilson.co.uk

drm/i915: Add fetch_and_zero() macro

A simple little macro to clear a pointer and return the old value. This
is useful for writing

value = *ptr;
if (!value)
return;

*ptr = 0;
...
free(value);

in a slightly more concise form:

value = fetch_and_zero(ptr);
if (!value)
return;

...
free(value);

with the idea that this establishes a pattern that may be extended for
atomic use (using xchg or cmpxchg) i.e. atomic_fetch_and_zero() and
similar to llist.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-10-git-send-email-chris@chris-wilson.co.uk

drm/i915: Create a VMA for an object

In many places, we wish to store the VMA in preference to the object
itself and so being able to create the persistent VMA is useful.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-9-git-send-email-chris@chris-wilson.co.uk

drm/i915: Always set the vma->pages

Previously, we would only set the vma->pages pointer for GGTT entries.
However, if we always set it, we can use it to prettify some code that
may want to access the backing store associated with the VMA (as
assigned to the VMA).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-8-git-send-email-chris@chris-wilson.co.uk

drm/i915: Remove redundant WARN_ON from __i915_add_request()

It's an outright programming error, so explode if it is ever hit.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-7-git-send-email-chris@chris-wilson.co.uk

drm/i915: Reduce i915_gem_objects to only show object information

No longer is knowing how much of the GTT (both mappable aperture and
beyond) relevant, and the output clutters the real information - that is
how many objects are allocated and bound (and by who) so that we can
quickly grasp if there is a leak.

v2: Relent, and rename pinned to indicate display only. Since the
display objects are semi-static and are of variable size, they are the
interesting objects to watch over time for aperture leaking. The other
pins are either static (such as the scratch page) or very short lived
(such as execbuf) and not part of the precious GGTT.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-6-git-send-email-chris@chris-wilson.co.uk

drm/i915: Focus debugfs/i915_gem_pinned to show only display pins

Only those objects pinned to the display have semi-permanent pins of a
global nature (other pins are transient within their local vm). Simplify
i915_gem_pinned to only show the pertinent information about the pinned
objects within the GGTT.

v2: i915_gem_gtt_info is still shared with debugfs/i915_gem_gtt,
rename i915_gem_pinned to i915_gem_pin_display to better reflect its
contents

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-5-git-send-email-chris@chris-wilson.co.uk

drm/i915: Remove inactive/active list from debugfs

These two files (i915_gem_active, i915_gem_inactive) no longer give
pertinent information since active/inactive tracking is per-vm and so we
need the information per-vm. They are obsolete so remove them.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-4-git-send-email-chris@chris-wilson.co.uk

drm/i915: Store the active context object on all engines upon error

With execlists, we have context objects everywhere, not just RCS. So
store them for post-mortem debugging. This also has a secondary effect
of removing one more unsafe list iteration with using preserved state
from the hanging request. And now we can cross-reference the request's
context state with that loaded by the GPU.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-3-git-send-email-chris@chris-wilson.co.uk

drm/i915: Reduce amount of duplicate buffer information captured on error

When capturing the error state, we do not need to know about every
address space - just those that are related to the error. We know which
context is active at the time, therefore we know which VM are implicated
in the error. We can then restrict the VM which we report to the
relevant subset.

v2: s/i/count_active/ (and similar)
    Rewrite label generation for "Buffers"

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-2-git-send-email-chris@chris-wilson.co.uk

drm/i915: Record the position of the start of the request

Not only does it make for good documentation and debugging aide, but it is
also vital for when we want to unwind requests - such as when throwing away
an incomplete request.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/1470414607-32453-2-git-send-email-arun.siluvery@linux.intel.com
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471254551-25805-1-git-send-email-chris@chris-wilson.co.uk

drm/i915: Show RPS autotuning thresholds along with waitboost

For convenience when debugging user issues show the autotuning
RPS parameters in debugfs/i915_rps_boost_info.

v2: Refine the presentation
v3: Style

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: fritsch@kodi.tv
Link: http://patchwork.freedesktop.org/patch/msgid/1471181336-27523-1-git-send-email-chris@chris-wilson.co.uk
Reviewed-by: David Weinehall <david.weinehall@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471250973-31277-1-git-send-email-chris@chris-wilson.co.uk

Merge remote-tracking branch 'airlied/drm-next' into drm-intel-next-queued

Backmerge because too many conflicts, and also we need to get at the
latest struct fence patches from Gustavo. Requested by Chris Wilson.

Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>

Merge tag 'drm-intel-next-2016-08-08' of git://anongit.freedesktop.org/drm-intel into drm-next

- refactor ddi buffer programming a bit (Ville)
- large-scale renaming to untangle naming in the gem code (Chris)
- rework vma/active tracking for accurately reaping idle mappings of shared
  objects (Chris)
- misc dp sst/mst probing corner case fixes (Ville)
- tons of cleanup&tunings all around in gem
- lockless (rcu-protected) request lookup, plus use it everywhere for
  non(b)locking waits (Chris)
- pipe crc debugfs fixes (Rodrigo)
- random fixes all over

* tag 'drm-intel-next-2016-08-08' of git://anongit.freedesktop.org/drm-intel: (222 commits)
  drm/i915: Update DRIVER_DATE to 20160808
  drm/i915: fix aliasing_ppgtt leak
  drm/i915: Update comment before i915_spin_request
  drm/i915: Use drm official vblank_no_hw_counter callback.
  drm/i915: Fix copy_to_user usage for pipe_crc
  Revert "drm/i915: Track active streams also for DP SST"
  drm/i915: fix WaInsertDummyPushConstPs
  drm/i915: Assert that the request hasn't been retired
  drm/i915: Repack fence tiling mode and stride into a single integer
  drm/i915: Document and reject invalid tiling modes
  drm/i915: Remove locking for get_tiling
  drm/i915: Remove pinned check from madvise ioctl
  drm/i915: Reduce locking inside swfinish ioctl
  drm/i915: Remove (struct_mutex) locking for busy-ioctl
  drm/i915: Remove (struct_mutex) locking for wait-ioctl
  drm/i915: Do a nonblocking wait first in pread/pwrite
  drm/i915: Remove unused no-shrinker-steal
  drm/i915: Tidy generation of the GTT mmap offset
  drm/i915/shrinker: Wait before acquiring struct_mutex under oom
  drm/i915: Simplify do_idling() (Ironlake vt-d w/a)
  ...

Merge tag 'topic/drm-misc-2016-08-12' of git://anongit.freedesktop.org/drm-intel into drm-next

- more fence destaging and cleanup (Gustavo&Sumit)
- DRIVER_LEGACY to untangle from DRIVER_MODESET
- drm_mm refactor (Chris)
- fbdev-less compile fies
- clipped plane src/dst rects (Ville)
- + a few mediatek patches that build on top of that (Bibby+Daniel)
- small stuff all over really

* tag 'topic/drm-misc-2016-08-12' of git://anongit.freedesktop.org/drm-intel: (43 commits)
  dma-buf/fence: kerneldoc: remove spurious section header
  dma-buf/fence: kerneldoc: remove unused struct members
  Revert "gpu: drm: omapdrm: dss-of: add missing of_node_put after calling of_parse_phandle"
  drm: Protect fb_defio in drivers with CONFIG_KMS_FBDEV_EMULATION
  drm/radeon|amgpu: Make fbdev emulation optional
  drm/vmwgfx: select CONFIG_FB
  drm: Remove superflous linux/fb.h includes
  drm/fb-helper: Add a dummy remove_conflicting_framebuffers
  dma-buf/sync_file: only enable fence signalling on poll()
  Documentation: add doc for sync_file_get_fence()
  dma-buf/sync_file: add sync_file_get_fence()
  dma-buf/sync_file: refactor fence storage in struct sync_file
  dma-buf/fence-array: add fence_is_array()
  drm/dp_helper: Rate limit timeout errors from drm_dp_i2c_do_msg()
  drm/dp_helper: Print first error received on failure in drm_dp_dpcd_access()
  drm: Add ratelimited versions of the DRM_DEBUG* macros
  drm: Make sure drm_vblank_no_hw_counter isn't abused
  drm/mediatek: Fix mtk_atomic_complete for runtime_pm
  drm/mediatek: plane: Use FB's format's cpp to compute x offset
  drm/mediatek: plane: Merge mtk_plane_enable into mtk_plane_atomic_update
  ...

Merge tag 'imx-drm-next-2016-08-12' of git://git.pengutronix.de/git/pza/linux into drm-next

imx-drm updates and encoder atomic_mode_set helper callback

- add pixel clock and DE polarity configuration from device tree
  using display timing bindings for parallel and LVDS output
- cleanup/remove trivial functions
- cleanup and fixes in preparation for capture support
- add atomic_mode_set helper and use it in imx-ldb - this is an
  alternative to the encoder mode_set callback that passes the
  crtc and connector state instead of just the mode. It allows
  drivers to get information from the attached connector without
  having to iterate over all connectors
- add drm_bridge support to imx-ldb, for bridges attached via LVDS

* tag 'imx-drm-next-2016-08-12' of git://git.pengutronix.de/git/pza/linux:
  drm/imx-ldb: Add support to drm-bridge
  drm/imx: imx-ldb: use encoder atomic_mode_set callback
  drm/atomic-helper: Add atomic_mode_set helper callback
  drm/imx: Remove imx_drm_handle_vblank()
  gpu: ipu-v3: Add missing IDMAC channel names
  gpu: ipu-v3: rename CSI client device
  gpu: ipu-v3: Fix IRT usage
  gpu: ipu-v3: Fix CSI data format for 16-bit media bus formats
  gpu: ipu-v3: set correct full sensor frame for PAL/NTSC
  gpu: ipu-v3: Add VDI input IDMAC channels
  gpu: ipu-v3: Add ipu_get_num()
  gpu: ipu-cpmem: Add ipu_cpmem_get_burstsize()
  gpu: ipu-cpmem: Add ipu_cpmem_set_uv_offset()
  drm/imx: Remove imx_drm_crtc_id()
  drm/imx: Remove imx_drm_crtc_vblank_get/_put()
  drm/imx: convey the pixelclk-active and de-active flags from DT to the ipu-di driver
  drm: add a helper function to extract 'de-active' and 'pixelclk-active' from DT

Merge tag 'mediatek-drm-next-2016-08-12' of git://git.pengutronix.de/git/pza/linux into drm-next

mediatek-drm maintainers and gamma correction

- add MAINTAINERS entry for mediatek-drm driver
- add support for AAL and GAMMA engines
- hook up gamma correction LUT
- add support for temporal dithering to OD and GAMMA engines

* tag 'mediatek-drm-next-2016-08-12' of git://git.pengutronix.de/git/pza/linux:
  drm/mediatek: set mt8173 dithering function
  drm/mediatek: Add gamma correction.
  drm/mediatek: Add GAMMA engine basic function
  drm/mediatek: Add AAL engine basic function
  drm: mediatek: add Maintainers entry for Mediatek DRM drivers

Merge branch 'drm-next-tilcdc-atomic' of https://github.com/jsarha/linux into drm-next

Please pull tilcdc atomic modeset support and some non critical fixes.

* 'drm-next-tilcdc-atomic' of https://github.com/jsarha/linux: (29 commits)
  drm/tilcdc: Change tilcdc_crtc_page_flip() to tilcdc_crtc_update_fb()
  drm/tilcdc: Remove unnecessary pm_runtime_get() and *_put() calls
  drm/tilcdc: Get rid of legacy dpms mechanism
  drm/tilcdc: Use drm_atomic_helper_resume/suspend()
  drm/tilcdc: Enable and disable interrupts in crtc start() and stop()
  drm/tilcdc: tfp410: Add atomic modeset helpers to connector funcs
  drm/tilcdc: tfp410: Set crtc panel info at init phase
  drm/tilcdc: panel: Add atomic modeset helpers to connector funcs
  drm/tilcdc: panel: Set crtc panel info at init phase
  drm/tilcdc: Remove tilcdc_verify_fb()
  drm/tilcdc: Remove obsolete crtc helper functions
  drm/tilcdc: Set DRIVER_ATOMIC and use atomic crtc helpers
  drm/tilcdc: Add drm_mode_config_reset() call to tilcdc_load()
  drm/tilcdc: Add atomic mode config funcs
  drm/tilcdc: Add tilcdc_crtc_atomic_check()
  drm/tilcdc: Add tilcdc_crtc_mode_set_nofb()
  drm/tilcdc: Initialize dummy primary plane from crtc init
  drm/tilcdc: Add dummy primary plane implementation
  drm/tilcdc: Make tilcdc_crtc_page_flip() work if crtc is not yet on
  drm/tilcdc: Make tilcdc_crtc_page_flip() public
  ...

Linux 4.8-rc2

Merge branch 'next' of git://git./linux/kernel/git/rzhang/linux

Pull thermal updates from Zhang Rui:

 - Fix a race condition when updating cooling device, which may lead to
   a situation where a thermal governor never updates the cooling
   device.  From Michele Di Giorgio.

 - Fix a zero division error when disabling the forced idle injection
   from the intel powerclamp.  From Petr Mladek.

 - Add suspend/resume callback for intel_pch_thermal thermal driver.
   From Srinivas Pandruvada.

 - Another two fixes for clocking cooling driver and hwmon sysfs I/F.
   From Michele Di Giorgio and Kuninori Morimoto.

[ Hmm.  That suspend/resume callback for intel_pch_thermal doesn't look
  like a fix, but I'm letting it slide..  - Linus ]

* 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/rzhang/linux:
  thermal: clock_cooling: Fix missing mutex_init()
  thermal: hwmon: EXPORT_SYMBOL_GPL for thermal hwmon sysfs
  thermal: fix race condition when updating cooling device
  thermal/powerclamp: Prevent division by zero when counting interval
  thermal: intel_pch_thermal: Add suspend/resume callback

Merge branch 'for-linus' of git://git./linux/kernel/git/gerg/m68knommu

Pull m68knommu fix from Greg Ungerer:
 "This contains only a single fix for a register corruption problem on
  certain types of m68k flat format binaries"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/gerg/m68knommu:
  m68knommu: fix user a5 register being overwritten

drm/i915: Unbind closed vma for i915_gem_object_unbind()

Closed vma are removed from the obj->vma_list so that they cannot be
found by userspace. However, this means that when forcibly unbinding an
object, we have to wait upon all rendering to that object first in order
for the closed, but active, vma to be reaped and their bindings removed.

Reported-by: Matthew Auld <matthew.auld@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=97343
Fixes: aa653a685d81 ("drm/i915: Be more careful when unbinding vma")
Fixes: 8a3b3d576c93 (" drm/i915: Convert non-blocking userptr waits...")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1471196681-30043-2-git-send-email-chris@chris-wilson.co.uk
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Tested-by: Matthew Auld <matthew.auld@intel.com>