Ville Syrjälä [Thu, 22 Oct 2015 12:35:00 +0000 (15:35 +0300)]
drm/i915: Add NEEDS_FORCEWAKE() checks for vlv/chv
Include an early NEEDS_FORCEWAKE() check for vlv and chv.
Hopefully that will avoid doing so many range checks in for many
register accesses (at least for all display registers).
Note that vlv already had the check in the write path since it shares
the gen6+ code for that.
Ville Syrjälä [Thu, 22 Oct 2015 12:34:59 +0000 (15:34 +0300)]
drm/i915: Respin vlv/chv reagister access to look more like SKL
Change the fw domain handling in the vlv/chv register read/write
functions to look more like the SKL code, ie. have a single
__force_wake_get() get call instead of multiple ones.
Ville Syrjälä [Thu, 22 Oct 2015 12:34:57 +0000 (15:34 +0300)]
drm/i915: Read FORCEWAKE registers with I915_READ_FW()
Change FORCEWAKE & co. reads for the error state to use I915_READ_FW().
Reading a FORCEWAKE register using a function that can frob forcewake
just seems wrong.
There is a check to skip grabbing the forcewake for accessing FORCEWAKE
in intel_uncore.c, but there's no such check for FORCEWAKE_MT. So no
idea what is currently happening with FORCEWAKE_MT reads. FORCEWAKE_VLV
is fortunately outside the forcewake range anyway, so no actual issue
with that one.
So let's just make the rule that you can't access FORCEWAKE registers with
the normal I915_READ() stuff, and we can drop the extra FORCEWAKE check
from NEEDS_FORCEWAKE(). While at it use NEEDS_FORCEWAKE() on BDW, where
it was skipped for whatever bikeshed reason that I've already forgotten.
Ville Syrjälä [Thu, 22 Oct 2015 12:34:56 +0000 (15:34 +0300)]
drm/i915: Turn __raw_i915_read8() & co. in to inline functions
There's no need for __raw_i915_read8() & co. to be macros, so make them
inline functions. To avoid typo mistakes generate the inline functions
using preprocessor templates.
We have a few users of the raw register acces functions outside
intel_uncore.c, so let's also move the functions into intel_drv.h.
While doing that switch I915_READ_FW() & co. to use the
__raw_i915_read() functions, and use the _FW macros everywhere
outside intel_uncore.c where we want to read registers without
grabbing forcewake and whatnot. The only exception is
i915_check_vgpu() which itself gets called from intel_uncore.c,
so using the __raw_i915_read stuff there seems appropriate.
v2: Squash in the intel_uncore.c->i915_drv.h move
Convert I915_READ_FW() to use __raw_i915_read(), and use
I915_READ_FW() outside of intel_uncore.c (Chris)
Tomas Elf [Mon, 19 Oct 2015 16:51:57 +0000 (17:51 +0100)]
drm/i915: Cope with request list state change during error state capture
Since we're not synchronizing the ring request list during error state capture
the request list state might change between the time the corresponding error
request list was allocated and dimensioned to the time when the ring request
list is actually captured into the error state. If this happens then do an
early exit and be aware that the captured error state might not be fully
reliable.
* v2:
- Chris Wilson: Removed WARN_ON from size check since having the error state
request list and the live driver request list diverge like this is a
legitimate behaviour.
- Tomas Elf: Removed update of num_request field since this made no sense. Just
exit and move on.
* v3:
- Chris Wilson: Removed error message at the point of early exit. The user is
not interested in any state changes happening during the error state capture,
only in the state that we're trying to capture at the point of the error.
Signed-off-by: Tomas Elf <tomas.elf@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Daniel Vetter [Fri, 23 Oct 2015 08:56:12 +0000 (10:56 +0200)]
drm/i915: shut up gen8+ SDE irq dmesg noise
We get tons of cases where the master interrupt handler apparently set
a bit, with the SDEIIR disagreeing. No idea what's going on there, but
it's consistent on gen8+, no one seems to care about it and it's
making CI results flaky.
Shut it up.
No idea what's going on here, but we've had fun with PCH interrupts
before:
drm/i915: also disable south interrupts when handling them
Note that there's a regression report in Bugzilla, and other
regression reports on the mailing lists keep croping up. But no ill
effects have ever been reported. But for paranoia still keep the
message at a debug level as a breadcrumb, just in case.
Tomas Elf [Mon, 19 Oct 2015 15:32:32 +0000 (16:32 +0100)]
drm/i915: Grab execlist spinlock to avoid post-reset concurrency issues.
Grab execlist lock when cleaning up execlist queues after GPU reset to avoid
concurrency problems between the context event interrupt handler and the reset
path immediately following a GPU reset.
* v2 (Chris Wilson):
Do execlist check and use simpler form of spinlock functions.
Signed-off-by: Tomas Elf <tomas.elf@intel.com> Reviewed-by: Dave Gordon <david.s.gordon@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Chris Wilson [Tue, 6 Oct 2015 10:53:11 +0000 (11:53 +0100)]
drm/i915: Add soft-pinning API for execbuffer
Userspace can pass in an offset that it presumes the object is located
at. The kernel will then do its utmost to fit the object into that
location. The assumption is that userspace is handling its own object
locations (for example along with full-ppgtt) and that the kernel will
rarely have to make space for the user's requests.
v2: Fix i915_gem_evict_range() (now evict_for_vma) to handle ordinary
and fixed objects within the same batch
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: "Daniel, Thomas" <thomas.daniel@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Matt Roper [Thu, 24 Sep 2015 22:53:17 +0000 (15:53 -0700)]
drm/i915: Don't set plane visible during HW readout if CRTC is off
We already ensure that pstate->visible = false when crtc->active = false
during runtime programming; make sure we follow the same logic when
reading out initial hardware state.
Signed-off-by: Matt Roper <matthew.d.roper@intel.com> Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Smoke-tested-by: Paulo Zanoni <przanoni@gmail.com> Link: http://patchwork.freedesktop.org/patch/59564/
Matt Roper [Thu, 24 Sep 2015 22:53:16 +0000 (15:53 -0700)]
drm/i915: Calculate ILK-style watermarks during atomic check (v3)
Calculate pipe watermarks during atomic calculation phase, based on the
contents of the atomic transaction's state structure. We still program
the watermarks at the same time we did before, but the computation now
happens much earlier.
While this patch isn't too exciting by itself, it paves the way for
future patches. The eventual goal (which will be realized in future
patches in this series) is to calculate multiple sets up watermark
values up front, and then program them at different times (pre- vs
post-vblank) on the platforms that need a two-step watermark update.
While we're at it, s/intel_compute_pipe_wm/ilk_compute_pipe_wm/ since
this function only applies to ILK-style watermarks and we have a
completely different function for SKL-style watermarks.
Note that the original code had a memcmp() in ilk_update_wm() to avoid
calling ilk_program_watermarks() if the watermarks hadn't changed. This
memcmp vanishes here, which means we may do some unnecessary result
generation and merging in cases where watermarks didn't change, but the
lower-level function ilk_write_wm_values already makes sure that we
don't actually try to program the watermark registers again.
v2: Squash a few commits from the original series together; no longer
leave pre-calculated wm's in a separate temporary structure since
it's easier to follow the logic if we just cut over to using the
pre-calculated values directly.
v3:
- Pass intel_crtc instead of drm_crtc to .compute_pipe_wm() entrypoint
and use intel_atomic_get_crtc_state() to avoid need for extra
casting. (Ander)
- Drop unused intel_check_crtc() function prototype. (Ander)
Signed-off-by: Matt Roper <matthew.d.roper@intel.com> Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Smoke-tested-by: Paulo Zanoni <przanoni@gmail.com> Link: http://patchwork.freedesktop.org/patch/60363/
Matt Roper [Thu, 24 Sep 2015 22:53:15 +0000 (15:53 -0700)]
drm/i915: Calculate pipe watermarks into CRTC state (v3)
A future patch will calculate these during the atomic 'check' phase
rather than at WM programming time, so let's store the watermark
values we're planning to use in the CRTC state; the values actually
active on the hardware remains in intel_crtc.
While we're at it, do some minor restructuring to keep ILK and SKL
values in a union.
v2: Don't move cxsr_allowed to state (Maarten)
v3: Only calculate watermarks in state. Still keep active watermarks in
intel_crtc itself. (Ville)
Signed-off-by: Matt Roper <matthew.d.roper@intel.com> Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Smoke-tested-by: Paulo Zanoni <przanoni@gmail.com> Link: http://patchwork.freedesktop.org/patch/59556/
Matt Roper [Thu, 24 Sep 2015 22:53:13 +0000 (15:53 -0700)]
drm/i915: Drop intel_update_sprite_watermarks
The only platform that still has an update_sprite_wm entrypoint is SKL;
on SKL, intel_update_sprite_watermarks just updates intel_plane->wm and
then performs a regular watermark update. However intel_plane->wm is
only used to update a couple fields in intel_wm_config, and those fields
are never used by the SKL code, so on SKL an update_sprite_wm is
effectively identical to an update_wm call. Since we're already
ensuring that the regular intel_update_wm is called any time we'd try to
call intel_update_sprite_watermarks, the whole call is redundant and can
be dropped.
Signed-off-by: Matt Roper <matthew.d.roper@intel.com> Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Smoke-tested-by: Paulo Zanoni <przanoni@gmail.com> Link: http://patchwork.freedesktop.org/patch/60372/
Matt Roper [Thu, 24 Sep 2015 22:53:12 +0000 (15:53 -0700)]
drm/i915/ivb: Move WaCxSRDisabledForSpriteScaling w/a to atomic check
Determine whether we need to apply this workaround at atomic check time
and just set a flag that will be used by the main watermark update
routine.
Moving this workaround into the atomic framework reduces
ilk_update_sprite_wm() to just a standard watermark update, so drop it
completely and just ensure that ilk_update_wm() is called whenever a
sprite plane is updated in a way that would affect watermarks.
Signed-off-by: Matt Roper <matthew.d.roper@intel.com> Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Smoke-tested-by: Paulo Zanoni <przanoni@gmail.com> Link: http://patchwork.freedesktop.org/patch/60367/
Alex Dai [Mon, 19 Oct 2015 23:10:54 +0000 (16:10 -0700)]
drm/i915/guc: Add GuC css header parser
The size / offset information of all firmware ingredients are
now caculated from header. Driver will validate the header and
rsa key size. If any component is out of boundary, driver will
reject the loading too.
v6: Clean up warnings from make docs
v5: Tidy up GuC titles in kernel/Doc
v4: Now using 'size_dw' for those defined in css_header
v3: 1) Move DOC to intel_guc_fwif.h right before css_header
definition. Add more comments.
2) Change 'size' to 'len' or 'length' to avoid confusion.
3) Add UOS_RSA_SCRATCH_MAX_COUNT according to BSpec. And
driver validate size of RSA key now.
4) Add fw component size/offset info to intel_guc_fw.
v2: Add indent into DOC to make fixed-width format rather than
change the tmpl.
v1: 1) guc_css_header is defined as __packed now
2) Add and correct GuC related topics in kernel/Doc
Signed-off-by: Alex Dai <yu.dai@intel.com> Reviewed-by: Dave Gordon <david.s.gordon@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Jani Nikula [Tue, 20 Oct 2015 12:22:02 +0000 (15:22 +0300)]
drm/i915: add helpers for platform specific revision id range checks
Revision checks are almost always accompanied by a platform check. (The
exceptions are platform specific code.) Add helpers to check for a
platform and a revision range: IS_SKL_REVID() and IS_BXT_REVID(). In
most places this simplifies and clarifies the code. It will be obvious
that revid macros are used for the correct platform.
This should make it easier to find all the revision checks for
workarounds for each platform, and make it easier to remove them once we
drop support for early hardware revisions.
This should also make it easier to differentiate between Skylake and
Kabylake revision checks when Kabylake support is added.
drm/i915: Do not wait for flips in intel_crtc_disable_noatomic.
intel_crtc_disable_noatomic is called from hw readout during init, resume and possibly reset.
During init it's too early to have a page flip queued, before suspending all page flips
should be finished and during hw reset all page flips should be removed.
It's a bug when there are pending flips here, complain with WARN_ON instead of handling it.
Tvrtko Ursulin [Tue, 20 Oct 2015 16:17:07 +0000 (17:17 +0100)]
drm/i915/skl: Assume no scaling is available when things are not as expected
Cdclk < crtc_clock is not allowed and suggests a different problem
elsewhere in the code.
It is more robust and safe to assume no scaling is possible in
this case with no other downsides since it will also WARN_ON_ONCE
so that this definitely gets noticed.
Call it an assert to help new platform bring-up in simulation.
v2: Better commit msg and use WARN_ON_ONCE to signify the unexpectedness.
v3: Move zero crtc_clock check under the warn. (Ville)
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Shobhit Kumar [Tue, 20 Oct 2015 12:43:12 +0000 (18:13 +0530)]
drm/i915/skl: If needed sanitize bios programmed cdclk
Especially in cases where pre-os does not enable display, cdclk might
not be in sane state. During sanitization initialize cdclk with maximum
value till we get dynamic cdclk support.
v2: Check if BIOS programmed correctly rather than always calling init
- Do validation of programmed cdctl and what it is expected
- Only do slk_init_cdclk if validation failed else reuse BIOS
programmed value
v3: Move the validation logic in a separate sanitize function (Ville)
v4: No need to check LCPLL after sanitize and use max_cdclk_freq instead
of hardcoded value (Ville)
Nick Hoath [Tue, 20 Oct 2015 09:23:52 +0000 (10:23 +0100)]
drm/i915: Break out common code from gen8_gt_irq_handler
Break out common code from gen8_gt_irq_handler and put it in to
an always inlined function. gcc optimises out the shift at compile
time. (Thomas Daniel/Daniel Vetter/Chris Wilson)
Matt Roper [Thu, 24 Sep 2015 22:53:11 +0000 (15:53 -0700)]
drm/i915/skl: Eliminate usage of pipe_wm_parameters from SKL-style WM (v4)
Just pull the info out of the state structures rather than staging
it in an additional set of structures. To make this more
straightforward, we change the signature of several internal WM
functions to take the crtc state as a parameter.
v2:
- Don't forget to skip cursor planes on a loop in the DDB allocation
function to match original behavior. (Ander)
- Change a use of intel_crtc->active to cstate->active. They should
be identical, but it's better to be consistent. (Ander)
- Rework more function signatures to pass states rather than crtc for
consistency. (Ander)
v3:
- Add missing "+ 1" to skl_wm_plane_id()'s 'overlay' case. (Maarten)
- Packed formats should pass '0' to drm_format_plane_cpp(), not 1.
(Maarten)
- Drop unwanted WARN_ON() for disabled planes when calculating data
rate for SKL. (Maarten)
v4:
- Don't include cursor plane in total relative data rate calculation;
we've already handled the cursor allocation earlier.
- Fix 'bytes_per_pixel' calculation braindamage. Somehow I hardcoded
the NV12 format as a parameter rather than the actual
fb->pixel_format, and even then still managed to get the format plane
wrong. (Ville)
- Use plane->state->fb rather than plane->fb in
skl_allocate_pipe_ddb(); the plane->fb pointer isn't updated until
after we've done our watermark recalculation, so it has stale
values. (Bob Paauwe)
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by(v3): Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Cc: Paauwe, Bob J <bob.j.paauwe@intel.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
References: http://lists.freedesktop.org/archives/intel-gfx/2015-September/077060.html
References: http://lists.freedesktop.org/archives/intel-gfx/2015-October/077721.html
Smoke-tested-by(v4): Paulo Zanoni <paulo.r.zanoni@intel.com> (SKL) Link: http://patchwork.freedesktop.org/patch/61968/
Chris Wilson [Wed, 14 Oct 2015 13:17:11 +0000 (14:17 +0100)]
drm/i915: Report context GTT size
Since the beginning we have conflated the size of the global GTT with
that of the per-process context sizes. In recent times (gen8+), those
are no longer the same where the global GTT is limited to 2/4GiB but the
per-process GTT may be anything up to 256TiB. Userspace knows nothing of
this discrepancy and outside of one or two hacks, uses the getaperture
ioctl to determine the maximum size it can use. Let's leave that as
reporting the global GTT and use the context reporting method to
describe the per-process value (which naturally fallsback to reporting
the aliasing or global on older platforms, so userspace can always use
this method where available).
Testcase: igt/gem_userptr_blits/minor-normal-sync
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=90065 Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Chris Wilson [Thu, 8 Oct 2015 12:39:54 +0000 (13:39 +0100)]
drm/i915: Map the ringbuffer using WB on LLC machines
If we have llc coherency, we can write directly into the ringbuffer
using ordinary cached writes rather than forcing WC access.
v2: An important consequence is that we can forgo the mappable request
for WB ringbuffers, allowing for many more simultaneous contexts.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Akash Goel <akash.goel@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Michel Thierry <michel.thierry@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drm/i915: Only call commit_planes when there are things to commit.
The atomic helpers set planes_changed on a crtc_state if there is
any plane_state bound to that crtc. If there's none and there is
no pipe update required the crtc has nothing to update, so vblank
evasion can be skipped.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drm/i915: Only commit active planes when updating planes during reset.
In the next commit commit_plane will no longer check if the crtc is active.
To prevent issues with legacy page flips the check should be performed inside
update_primary_planes.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drm/i915: Update legacy primary state outside the commit hook, v2.
This should allow not running plane commit when the crtc is off.
While the atomic helpers update those, crtc->x/y is only updated
during modesets, and primary plane is updated after this function
returns.
Unfortunately non-atomic watermarks and fbc still depend on this
state inside i915, so it has to be kept in sync.
Changes since v1:
- Add comment that the legacy state is updated for fbc.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drm/i915: Make prepare_plane_fb fully interruptible.
Now that we agreed on not preserving framebuffers pinning is finally
allowed to fail because of signals. Use this to make pinning
and acquire the mutex in an interruptible way too.
Unpinning is still uninterruptible, because it happens as a cleanup
of old state, or undoing pins after one of the pins failed.
The intel_pin_and_fence_fb_obj in page_flip will also wait interruptibly,
and can be aborted now.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Reviewed-by: Ander Conselvan de Oliveira <conselvan2@gmail.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drm/i915: Make plane fb tracking work correctly, v2.
atomic->disabled_planes is a hack that had to exist because
prepare_fb was only called when a new fb was set. This messed
up fb tracking in some circumstances like aborts from
interruptible waits. As a result interruptible waiting in
prepare_plane_fb was forbidden, but other errors could still
cause frontbuffer tracking to be messed up.
Now that prepare_fb is always called, this hack is no longer
required and prepare_fb may fail without consequences.
Changes since v1:
- Clean up a few fb tracking warnings by changing plane->fb to
plane->state->fb.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> Reviewed-by: Ander Conselvan de Oliveira <conselvan2@gmail.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Animesh Manna [Tue, 29 Sep 2015 05:31:59 +0000 (11:01 +0530)]
drm/i915/skl: Making DC6 entry is the last call in suspend flow.
Mmio register access after dc6/dc5 entry is not allowed when
DC6 power states are enabled according to bspec (bspec-id 0527),
so enabling dc6 as the last call in suspend flow.
Addtional note from Imre:
Currently we keep DC6 enabled during modesets and DPAUX transfers, which
is not allowed according to the specification. This can lead at least to
PLL locking failures, DPAUX timeouts and prevent deeper package power
states (PC9/10). Fix this for now by enabling DC6 only when we know the
above events (modeset, DPAUX) can't happen.
This a temporary solution as some issues are still unsolved as described
in [1] and [2], we'll address those as a follow-up.
if forgotten to take into account there case where we have a
GLOBAL_BIND before a LOCAL_BIND. This patch here fixes that.
v2: Pimp commit message and revert the partial fix.
v3: Split into two functions to specialize on aliasing_ppgtt y/n.
v4: WARN_ON for paranoia in the init sequence, since the ggtt probe
and aliasing ppgtt setup are far apart.
v5: Style nits.
Cc: Chris Wilson <chris@chris-wilson.co.uk> Cc: Michel Thierry <michel.thierry@intel.com> Cc: Mika Kuoppala <mika.kuoppala@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com> Link: http://mid.gmane.org/1444911781-32607-1-git-send-email-daniel.vetter@ffwll.ch Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Bob Paauwe [Thu, 27 Aug 2015 20:46:30 +0000 (13:46 -0700)]
drm/i915/skl: Enable pipe gamma for sprite planes.
Since SKL has universal planes, we should configure the sprite planes
and the primary plane the same. For the primary plane we do enable
the pipe gamma on the plane so do the same for the non-primary planes.
Without this, the pipe CRC values will be different for something
displayed on the primary plane and something displayed on a sprite
plane when the ARGB8888 format is used.
Signed-off-by: Bob Paauwe <bob.j.paauwe@intel.com> Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Bob Paauwe [Mon, 31 Aug 2015 21:03:30 +0000 (14:03 -0700)]
drm/i915/skl+: Enable pipe CSC on cursor planes. (v2)
Extend this to SKL and BXT as it's needed for these platforms as well.
v2: Change if condition to HAS_DDI() instead of listing each platform Signed-off-by: Bob Paauwe <bob.j.paauwe@intel.com> Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Mika Kuoppala [Mon, 12 Oct 2015 10:20:59 +0000 (13:20 +0300)]
drm/i915: Move skl/bxt gt specific workarounds to ring init
Some registers are, naturally, lost in gpu reset/suspend cycle.
And some registers, for example in display domain, are not subject
to gpu reset so they retain their contents.
As hang recovery triggers a reset, recoverable gpu hang can currently
flush out essential workarounds and cause havoc later on.
When register GEN8_GARBNTL is missing the WaEnableGapsTsvCreditFix:skl,
it can cause random system hangs [1]. This workaround was added in:
commit 245d96670d26 ("drm/i915:skl: Add WaEnableGapsTsvCreditFix")
But another set of system hangs were observed and the failure pattern
indicated that there was random gpu hang preceding the system hang [2].
This lead to the realization that we lose this workaround and BDW_SCRATCH1
on reset.
Add these workarounds setup in display init to skl/bxt ring init
where LRI workarounds are also setup. This way their setup is not
dependent on display side init.
References: [1] https://bugs.freedesktop.org/show_bug.cgi?id=90854
References: [2] https://bugs.freedesktop.org/show_bug.cgi?id=92315 Reported-by: Tomi Sarvela <tomix.p.sarvela@intel.com> Cc: Tomi Sarvela <tomix.p.sarvela@intel.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Tested-by: Tomi Sarvela <tomix.p.sarvela@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Chris Wilson [Fri, 9 Oct 2015 13:11:27 +0000 (14:11 +0100)]
drm/i915: Drop i915_gem_obj_is_pinned() from set-cache-level
Since the remove of the pin-ioctl, we only care about not changing the
cache level on buffers pinned to the hardware as indicated by
obj->pin_display. By knowing that only objects pinned to the hardware
will have an elevated vma->pin_count, so we can coallesce many of the
linear walks over the obj->vma_list.
v2: Try and retrospectively add comments explaining the steps in
rebinding the active VMA.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
With these reverts, SKL finally stops failing every single FBC test
with FIFO underrun error messages. After some brief testing, it also
seems that this commit prevents the machine from completely freezing
when we run igt/kms_fbc_crc (see fd.o #92355).
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92355 Cc: Matt Roper <matthew.d.roper@intel.com> Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Ville Syrjälä [Fri, 18 Sep 2015 17:03:42 +0000 (20:03 +0300)]
drm/i915: s/PIPE_FRMCOUNT_GM45/PIPE_FRMCOUNT_G4X/ etc.
The PIPE_FRMCOUNT_GM45 and PIPE_FLIPCOUNT_GM45 names have bothered me
for a long time. The work equally well for ELK and onwards, so let's
s/GM45/G4X/.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Chris Wilson [Sat, 10 Oct 2015 09:44:32 +0000 (10:44 +0100)]
drm/i915: Hold dev->event_lock whilst inspecting intel_crtc->unpin_work
We should serialise access to the intel_crtc->unpin_work through the
dev->event_lock spinlock. It should not be possible for it to disappear
without severe error as the mmio_flip worker has not tagged the
unpin_work pending flip-completion. Similarly if the error exists, just
taking the unpin_work whilst holding the spinlock and then using it
unserialised just masks the race. (It is supposed to be valid as the
unpin_work exists until the flip completion interrupt which should not
fire until we flush the mmio writes to update the display base which is
the last time we access the unpin_work from the kthread.)
References: https://bugs.freedesktop.org/show_bug.cgi?id=92335 Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Williams, Dan J [Mon, 12 Oct 2015 21:12:57 +0000 (21:12 +0000)]
i915: switch from acpi_os_ioremap to memremap
i915 expects the OpRegion to be cached (i.e. not __iomem), so explicitly
map it with memremap rather than the implied cache setting of
acpi_os_ioremap().
Cc: Daniel Vetter <daniel.vetter@intel.com> Cc: Jani Nikula <jani.nikula@linux.intel.com> Cc: intel-gfx@lists.freedesktop.org Cc: David Airlie <airlied@linux.ie> Cc: dri-devel@lists.freedesktop.org Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Lukas Wunner [Mon, 12 Oct 2015 08:57:43 +0000 (10:57 +0200)]
drm/i915: Drop unnecessary #include <linux/vga_switcheroo.h>
Commit 599bbb9de0fe ("drm/i915: i915 cannot provide switcher services.")
removed all remaining vga_switcheroo symbols from intel_acpi.c but left
the include. Drop it.
Signed-off-by: Lukas Wunner <lukas@wunner.de> Reviewed-by: Jani Nikula <jani.nikula@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Matt Roper [Thu, 8 Oct 2015 22:28:25 +0000 (15:28 -0700)]
drm/i915: Partial revert of atomic watermark series
It's been reported that the atomic watermark series triggers some
regressions on SKL, which we haven't been able to track down yet. Let's
temporarily revert these patches while we track down the root cause.
This commit squashes the reverts of: 76305b1 drm/i915: Calculate watermark configuration during atomic check (v2) a4611e4 drm/i915: Don't set plane visible during HW readout if CRTC is off a28170f drm/i915: Calculate ILK-style watermarks during atomic check (v3) de4a9f8 drm/i915: Calculate pipe watermarks into CRTC state (v3) de165e0 drm/i915: Refactor ilk_update_wm (v3)
Reference: http://lists.freedesktop.org/archives/intel-gfx/2015-October/077190.html Cc: "Zanoni, Paulo R" <paulo.r.zanoni@intel.com> Cc: "Vetter, Daniel" <daniel.vetter@intel.com> Signed-off-by: Matt Roper <matthew.d.roper@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Tomas Elf [Thu, 8 Oct 2015 18:31:33 +0000 (19:31 +0100)]
drm/i915: Early exit from semaphore_waits_for for execlist mode.
When submitting semaphores in execlist mode the hang checker crashes in this
function because it is only runnable in ring submission mode. The reason this
is of particular interest to the TDR patch series is because we use semaphores
as a mean to induce hangs during testing (which is the recommended way to
induce hangs for gen8+). It's not clear how this is supposed to work in
execlist mode since:
1. This function requires a ring buffer.
2. Retrieving a ring buffer in execlist mode requires us to retrieve the
corresponding context, which we get from a request.
3. Retieving a request from the hang checker is not straight-forward since that
requires us to grab the struct_mutex in order to synchronize against the
request retirement thread.
4. Grabbing the struct_mutex from the hang checker is nothing that we will do
since that puts us at risk of deadlock since a hung thread might be holding the
struct_mutex already.
Therefore it's not obvious how we're supposed to deal with this. For now, we're
doing an early exit from this function, which avoids any kernel panic situation
when running our own internal TDR ULT.
* v2: (Chris Wilson)
Turned the execlist mode check into a ringbuffer NULL check to make it more
submission mode agnostic and less of a layering violation.
Signed-off-by: Tomas Elf <tomas.elf@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drm/i915: Clean up associated VMAs on context destruction
Introduced a wrong assumption that all contexts have a ppgtt
instance. This is not true when full PPGTT is not active so
remove the WARN_ON_ONCE from the context cleanup code.
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Michel Thierry <michel.thierry@intel.com> Reviewed-by: Michel Thierry <michel.thierry@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Ville Syrjälä [Thu, 8 Oct 2015 09:08:20 +0000 (12:08 +0300)]
drm/i915: Determine the stolen memory base address on gen2
There isn't an explicit stolen memory base register on gen2.
Some old comment in the i915 code suggests we should get it via
max_low_pfn_mapped, but that's clearly a bad idea on my MGM.
That makes max_low_pfn_mapped = 1f6f0000, so assuming our stolen memory
would start there would place it on top of some ACPI memory regions.
So not a good idea as already stated.
The 9MB region after the ACPI regions at 0x1f700000 however looks
promising given that the macine reports the stolen memory size to be
8MB. Looking at the PGTBL_CTL register, the GTT entries are at offset
0x1fee00000, and given that the GTT entries occupy 128KB, it looks like
the stolen memory could start at 0x1f700000 and the GTT entries would
occupy the last 128KB of the stolen memory.
After some more digging through chipset documentation, I've determined
the BIOS first allocates space for something called TSEG (something to
do with SMM) from the top of memory, and then it allocates the graphics
stolen memory below that. Accordind to the chipset documentation TSEG
has a fixed size of 1MB on 855. So that explains the top 1MB in the
e820 region. And it also confirms that the GTT entries are in fact at
the end of the the stolen memory region.
Derive the stolen memory base address on gen2 the same as the BIOS does
(TOM-TSEG_SIZE-stolen_size). There are a few differences between the
registers on various gen2 chipsets, so a few different codepaths are
required.
865G is again bit more special since it seems to support enough memory
to hit 4GB address space issues. This means the PCI allocations will
also affect the location of the stolen memory. Fortunately there
appears to be the TOUD register which may give us the correct answer
directly. But the chipset docs are a bit unclear, so I'm not 100%
sure that the graphics stolen memory is always the last thing the
BIOS steals. Someone would need to verify it on a real system.
I tested this on the my 830 and 855 machines, and so far everything
looks peachy.
v2: Rewrite to use the TOM-TSEG_SIZE-stolen_size and TOUD methods
v3: Fix TSEG size for 830
v4: Add missing 'else' (Chris)
Tested-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Paulo Zanoni [Thu, 1 Oct 2015 22:57:12 +0000 (19:57 -0300)]
drm/i915: fix FBC buffer size checks
According to my experiments (and later confirmation from the hardware
developers), the maximum sizes mentioned in the specification delimit
how far in the buffer the hardware tracking can go. And the hardware
calculates the size based on the plane address we provide - and the
provided plane address might not be the real x:0,y:0 point due to the
compute_page_offset() function.
On platforms that do the x/y offset adjustment trick it will be really
hard to reproduce a bug, but on the current SKL we can reproduce the
bug with igt/kms_frontbuffer_tracking/fbc-farfromfence. With this
patch, we'll go from "CRC assertion failure" to "FBC unexpectedly
disabled", which is still a failure on the test suite but is not a
perceived user bug - you will just not save as much power as you could
if FBC is disabled.
v2, rewrite patch after clarification from the Hadware guys:
- Rename function so it's clear what the check is for.
- Use the new intel_fbc_get_plane_source_sizes() function in order
to get the proper sizes as seen by FBC.
v3:
- Rebase after the s/sizes/size/ on the previous patch.
- Adjust comment wording (Ville).
- s/used_/effective_/ (Ville).
Testcase: igt/kms_frontbuffer_tracking/fbc-farfromfence (SKL) Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Paulo Zanoni [Thu, 1 Oct 2015 22:55:57 +0000 (19:55 -0300)]
drm/i915: fix CFB size calculation
We were considering the whole framebuffer height, but the spec says we
should only consider the active display height size. There were still
some unclear questions based on the spec, but the hardware guys
clarified them for us. According to them:
- CFB size = CFB stride * Number of lines FBC writes to CFB
- CFB stride = plane stride / compression limit
- Number of lines FBC writes to CFB = MIN(plane source height, maximum
number of lines FBC writes to CFB)
- Plane source height =
- pipe source height (PIPE_SRCSZ register) (before SKL)
- plane size register height (PLANE_SIZE register) (SKL+)
- Maximum number of lines FBC writes to CFB =
- plane source height (before HSW)
- 2048 (HSW+)
For the plane source height, I could just have made our code do
I915_READ() in order to be more future proof, but since it's not cool
to do register reads I decided to just recalculate the values we use
when we actually write to those registers.
With this patch, depending on your machine configuration, a lot of the
kms_frontbuffer_tracking subtests that used to result in a SKIP due to
not enough stolen memory still start resulting in a PASS.
v2: Use the clipped src size instead of pipe_src_h (Ville).
v3: Use the appropriate information provided by the hardware guys.
v4: Bikesheds: s/sizes/size/, s/fb_cpp/cpp/ (Ville).
v5: - Don't use crtc->config->pipe_src_x for BDW- (Ville).
- Fix the register name written in the comment.
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com> Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Paulo Zanoni [Wed, 30 Sep 2015 20:05:43 +0000 (17:05 -0300)]
drm/i915: remove pre-atomic check from SKL update_primary_plane
The comment suggests the check was there for some non-fully-atomic
case, and I couldn't find a case where we wouldn't correctly
initialize plane_state, so remove the check.
Let's leave a WARN there just in case.
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com> Acked-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Paulo Zanoni [Wed, 23 Sep 2015 15:52:23 +0000 (12:52 -0300)]
drm/i915: don't allocate fbcon from stolen memory if it's too big
Technology has evolved and now we have eDP panels with 3200x1800
resolution. In the meantime, the BIOS guys didn't change the default
32mb for stolen memory. On top of that, we can't assume our users will
be able to increase the default stolen memory size to more than 32mb -
I'm not even sure all BIOSes allow that.
So just the fbcon buffer alone eats 22mb of my stolen memroy, and due
to the BDW/SKL restriction of not using the last 8mb of stolen memory,
all that's left for FBC is 2mb! Since fbcon is not the coolest feature
ever, I think it's better to save our precious stolen resource to FBC
and the other guys.
On the other hand, we really want to use as much stolen memory as
possible, since on some older systems the stolen memory may be a
considerable percentage of the total available memory.
This patch tries to achieve a little balance using a simple heuristic:
if the fbcon wants more than half of the available stolen memory,
don't use stolen memory in order to leave some for FBC and the other
features.
The long term plan should be to implement a way to set priorities for
stolen memory allocation and then evict low priority users when the
high priority ones need the memory. While we still don't have that,
let's try to make FBC usable with the simple solution.
Cc: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com> Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
It results on a deadlock on platforms where we need to (at least
partially) re-init hpd interrupts from power domain code, since
->hot_plug might again grab a power well reference (to do edid/dp_aux
transactions. At least chv is affected.
Reported-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
References: http://mid.gmane.org/20151008133548.GX26517@intel.com Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Sudip Mukherjee [Thu, 8 Oct 2015 13:57:59 +0000 (19:27 +0530)]
drm/i915: use error path
Use goto to handle the error path to avoid duplicating the same code. In
the error path intel_dig_port is the last one to be released as it was
the first one to be allocated and ideally the error path should be the
reverse of the execution path.
Cc: Daniel Vetter <daniel.vetter@intel.com> Cc: Jani Nikula <jani.nikula@linux.intel.com> Signed-off-by: Sudip Mukherjee <sudip@vectorindia.org> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Add the dev parameter for the functions i915_enable_asle_pipestat() and
i915_reset_and_wakeup() to the kernel-doc to fix the following warnings:
.//drivers/gpu/drm/i915/i915_irq.c:586: warning: No description found for parameter 'dev'
.//drivers/gpu/drm/i915/i915_irq.c:2400: warning: No description found for parameter 'dev'
Signed-off-by: Javier Martinez Canillas <javier@osg.samsung.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Francisco Jerez [Wed, 7 Oct 2015 11:44:02 +0000 (14:44 +0300)]
drm/i915: Hook up ring workaround writes at context creation time on Gen6-7.
intel_rcs_ctx_init() emits all workaround register writes on the list
to the ring, in addition to calling i915_gem_render_state_init(). The
workaround list is currently empty on Gen6-7 so this shouldn't cause
any functional changes.
Signed-off-by: Francisco Jerez <currojerez@riseup.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Francisco Jerez [Wed, 7 Oct 2015 11:44:01 +0000 (14:44 +0300)]
drm/i915: Don't warn if the workaround list is empty.
It's not an error for the workaround list to be empty if no
workarounds are needed. This will avoid spamming the logs
unnecessarily on Gen6 after the workaround list is hooked up on
pre-Gen8 hardware by the following commits.
Signed-off-by: Francisco Jerez <currojerez@riseup.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drm/i915: Move golden context init into ->init_context
I've shuffled around per-ctx init code a bit for legacy contexts but
accidentally dropped the render state init call on gen6/7. Resurrect
it.
Reported-by: Francisco Jerez <currojerez@riseup.net> Cc: Francisco Jerez <currojerez@riseup.net> Cc: Dave Gordon <david.s.gordon@intel.com> Cc: Thomas Daniel <thomas.daniel@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com> Reviewed-by: Francisco Jerez <currojerez@riseup.net> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Jani Nikula <jani.nikula@intel.com> Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
[danvet: Appease gcc and remove the unused variable.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Jani Nikula <jani.nikula@intel.com> Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Chris Wilson [Tue, 6 Oct 2015 10:39:55 +0000 (11:39 +0100)]
drm/i915: Kill DRI1 cliprects
Passing cliprects into the kernel for it to re-execute the batch buffer
with different CMD_DRAWRECT died out long ago. As DRI1 support has been
removed from the kernel, we can now simply reject any execbuf trying to
use this "feature".
To keep Daniel happy with the prospect of being able to reuse these
fields in the next decade, continue to ensure that current userspace is
not passing garbage in through the dead fields.
v2: Fix the cliprects_ptr check
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Dave Gordon <david.s.gordon@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Chris Wilson [Thu, 1 Oct 2015 11:18:29 +0000 (12:18 +0100)]
drm/i915: Avoid GPU stalls from kswapd
Exclude active GPU pages from the purview of the background shrinker
(kswapd), as these cause uncontrollable GPU stalls. Given that the
shrinker is rerun until the freelists are satisfied, we should have
opportunity in subsequent passes to recover the pages once idle. If the
machine does run out of memory entirely, we have the forced idling in the
oom-notifier as a means of releasing all the pages we can before an oom
is prematurely executed.
Note that this relies upon an up-front retire_requests to keep the
inactive list in shape, which was added in a previous patch, mostly as
execlist ctx pinning band-aids.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
[danvet: Add note about retire_requests.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Chris Wilson [Thu, 1 Oct 2015 11:18:27 +0000 (12:18 +0100)]
drm/i915: During shrink_all we only need to idle the GPU
We can forgo an evict-everything here as the shrinker operation itself
will unbind any vma as required. If we explicitly idle the GPU through a
switch to the default context, we not only create a request in an
illegal context (e.g. whilst shrinking during execbuf with a request
already allocated), but switching to the default context will not free
up the memory backing the active contexts - unless in the unlikely
situation that context had already been closed (and just kept arrive by
being the current context). The saving is near zero and the danger real.
To compensate for the loss of the forced retire, add a couple of
retire-requests to i915_gem_shirnk() - this should help free up any
transitive cache from the requests.
Note that the second retire_requests is for the benefit of the
hand-rolled execlist ctx active tracking: We need to manually kick
requests to get those unpinned again. Once that's fixed we can try to
remove this again.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Add summary of why we need a pile of retire_requests.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Chris Wilson [Thu, 1 Oct 2015 11:18:26 +0000 (12:18 +0100)]
drm/i915: Add a tracepoint for the shrinker
Often it is very useful to know why we suddenly purge vast tracts of
memory and surprisingly up until now we didn't even have a tracepoint
for when we shrink our memory.
Note that there are slab_start/end tracepoints already, but those
don't cover the internal recursion when we directly call into our
shrinker code. Hence a separate tracepoint seems justified. Also note
that we don't really need a separate tracepoint for the actual amount
of pages freed since we already have an unbind tracpoint for that.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Add a note that there's also slab_start/end and why they're
insufficient.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Libin Yang [Thu, 1 Oct 2015 09:01:09 +0000 (17:01 +0800)]
drm/i915: DocBook add i915_component.h support
Add the item of i915_component.h in DocBook and add the DOC for
i915_component.h. Explain the struct i915_audio_component_ops and
struct i915_audio_component_audio_ops usage.
Signed-off-by: Libin Yang <libin.yang@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Chris Wilson [Thu, 1 Oct 2015 11:34:47 +0000 (12:34 +0100)]
drm/i915: Use a task to cancel the userptr on invalidate_range
Whilst discussing possible ways to trigger an invalidate_range on a
userptr with an aliased GGTT mmapping (and so cause a struct_mutex
deadlock), the conclusion is that we can, and we must, prevent any
possible deadlock by avoiding taking the mutex at all during
invalidate_range. This has numerous advantages all of which stem from
avoid the sleeping function from inside the unknown context. In
particular, it simplifies the invalidate_range because we no longer
have to juggle the spinlock/mutex and can just hold the spinlock
for the entire walk. To compensate, we have to make get_pages a bit more
complicated in order to serialise with a pending cancel_userptr worker.
As we hold the struct_mutex, we have no choice but to return EAGAIN and
hope that the worker is then flushed before we retry after reacquiring
the struct_mutex.
The important caveat is that the invalidate_range itself is no longer
synchronous. There exists a small but definite period in time in which
the old PTE's page remain accessible via the GPU. Note however that the
physical pages themselves are not invalidated by the mmu_notifier, just
the CPU view of the address space. The impact should be limited to a
delay in pages being flushed, rather than a possibility of writing to
the wrong pages. The only race condition that this worsens is remapping
an userptr active on the GPU where fresh work may still reference the
old pages due to struct_mutex contention. Given that userspace is racing
with the GPU, it is fair to say that the results are undefined.
v2: Only queue (and importantly only take one refcnt) the worker once.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Michał Winiarski <michal.winiarski@intel.com> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Chris Wilson [Thu, 1 Oct 2015 11:34:46 +0000 (12:34 +0100)]
drm/i915: Fix userptr deadlock with aliased GTT mmappings
Michał Winiarski found a really evil way to trigger a struct_mutex
deadlock with userptr. He found that if he allocated a userptr bo and
then GTT mmaped another bo, or even itself, at the same address as the
userptr using MAP_FIXED, he could then cause a deadlock any time we then
had to invalidate the GTT mmappings (so at will). Tvrtko then found by
repeatedly allocating GTT mmappings he could alias with an old userptr
mmap and also trigger the deadlock.
To counter act the deadlock, we make the observation that we only need
to take the struct_mutex if the object has any pages to revoke, and that
before userspace can alias with the userptr address space, it must have
invalidated the userptr->pages. Thus if we can check for those pages
outside of the struct_mutex, we can avoid the deadlock. To do so we
introduce a separate flag for userptr objects that we can inspect from
the mmu-notifier underneath its spinlock.
The patch makes one eye-catching change. That is the removal serial=0
after detecting a to-be-freed object inside the invalidate walker. I
felt setting serial=0 was a questionable pessimisation: it denies us the
chance to reuse the current iterator for the next loop (before it is
freed) and being explicit makes the reader question the validity of the
locking (since the object-free race could occur elsewhere). The
serialisation of the iterator is through the spinlock, if the object is
freed before the next loop then the notifier.serial will be incremented
and we start the walk from the beginning as we detect the invalid cache.
To try and tame the error paths and interactions with the userptr->active
flag, we have to do a fair amount of rearranging of get_pages_userptr().
v2: Grammar fixes
v3: Reorder set-active so that it is only set when obj->pages is set
(and so needs cancellation). Only the order of setting obj->pages and
the active-flag is crucial. Calling gup after invalidate-range begin
means the userptr sees the new set of backing storage (and so will not
need to invalidate its new pages), but we have to be careful not to set
the active-flag prior to successfully establishing obj->pages.
v4: Take the active->flag early so we know in the mmu-notifier when we
have to cancel a pending gup-worker.
v5: Rearrange the error path so that is not so convoluted
v6: Set pinned to 0 when negative before calling release_pages()
Reported-by: Michał Winiarski <michal.winiarski@intel.com>
Testcase: igt/gem_userptr_blits/map-fixed* Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Michał Winiarski <michal.winiarski@intel.com> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: stable@vger.kernel.org Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Chris Wilson [Thu, 1 Oct 2015 11:34:45 +0000 (12:34 +0100)]
drm/i915: Only update the current userptr worker
The userptr worker allows for a slight race condition where upon there
may two or more threads calling get_user_pages for the same object. When
we have the array of pages, then we serialise the update of the object.
However, the worker should only overwrite the obj->userptr.work pointer
if and only if it is the active one. Currently we clear it for a
secondary worker with the effect that we may rarely force a second
lookup.
v2: Rebase and rename a variable to avoid 80cols
v3: Mention v2
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Michel Thierry [Fri, 2 Oct 2015 13:16:53 +0000 (14:16 +0100)]
drm/i915: prevent out of range pt in the PDE macros (take 3)
We tried to fix this in commit fdc454c1484a ("drm/i915: Prevent out of
range pt in gen6_for_each_pde").
But the static analyzer still complains that, just before we break due
to "iter < I915_PDES", we do "pt = (pd)->page_table[iter]" with an
iter value that is bigger than I915_PDES. Of course, this isn't really
a problem since no one uses pt outside the macro. Still, every single
new usage of the macro will create a new issue for us to mark as a
false positive.
Also, Paulo re-started the discussion a while ago [1], but didn't end up
implemented.
In order to "solve" this "problem", this patch takes the ideas from
Chris and Dave, but that check would change the desired behavior of the
code, because the object (for example pdp->page_directory[iter]) can be
null during init/alloc, and C would take this as false, breaking the for
loop immediately.
This has been already verified with "static analysis tools".
v2: Make it a single statement, while preventing the common subexpression
elimination (Chris)
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Cc: Dave Gordon <david.s.gordon@intel.com> Signed-off-by: Michel Thierry <michel.thierry@intel.com> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Tvrtko Ursulin [Mon, 5 Oct 2015 12:26:36 +0000 (13:26 +0100)]
drm/i915: Clean up associated VMAs on context destruction
Prevent leaking VMAs and PPGTT VMs when objects are imported
via flink.
Scenario is that any VMAs created by the importer will be left
dangling after the importer exits, or destroys the PPGTT context
with which they are associated.
This is caused by object destruction not running when the
importer closes the buffer object handle due the reference held
by the exporter. This also leaks the VM since the VMA has a
reference on it.
In practice these leaks can be observed by stopping and starting
the X server on a kernel with fbcon compiled in. Every time
X server exits another VMA will be leaked against the fbcon's
frame buffer object.
Also on systems where flink buffer sharing is used extensively,
like Android, this leak has even more serious consequences.
This version is takes a general approach from the earlier work
by Rafael Barbalho (drm/i915: Clean-up PPGTT on context
destruction) and tries to incorporate the subsequent discussion
between Chris Wilson and Daniel Vetter.
v2:
Removed immediate cleanup on object retire - it was causing a
recursive VMA unbind via i915_gem_object_wait_rendering. And
it is in fact not even needed since by definition context
cleanup worker runs only after the last context reference has
been dropped, hence all VMAs against the VM belonging to the
context are already on the inactive list.
v3:
Previous version could deadlock since VMA unbind waits on any
rendering on an object to complete. Objects can be busy in a
different VM which would mean that the cleanup loop would do
the wait with the struct mutex held.
This is an even simpler approach where we just unbind VMAs
without waiting since we know all VMAs belonging to this VM
are idle, and there is nothing in flight, at the point
context destructor runs.
v4:
Double underscore prefix for __915_vma_unbind_no_wait and a
commit message typo fix. (Michel Thierry)
Note that this is just a partial/interim fix since we have a bit a
fundamental issue with cleaning up, e.g.
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Testcase: igt/gem_ppgtt.c/flink-and-exit-vma-leak Reviewed-by: Michel Thierry <michel.thierry@intel.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: Chris Wilson <chris@chris-wilson.co.uk> Cc: Rafael Barbalho <rafael.barbalho@intel.com> Cc: Michel Thierry <michel.thierry@intel.com>
[danvet: Add a note that this isn't everything.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The link training functions had confusing names. The start function
actually does the clock recovery phase of the link training, and the
complete function does the channel equalization. So call them that
instead. Also, every call to intel_dp_start_link_train() was followed
by a call to intel_dp_complete_link_train(), so add a new start
function that calls clock_recory and channel_equalization.
Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
projects / linux-beck.git / log