We assume that the panel is permenantly connected and that the EDID data
is consistent from boot, so simply cache the whole EDID for the panel.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
During heavy aperture thrashing we may be forced to wait upon several active
objects during eviction. The active list may be the last reference to
these objects and so the action of waiting upon one of them may cause
another to be freed (and itself unbound). To prevent the object
disappearing underneath us, we need to acquire and hold a reference
whilst unbinding.
This should fix the reported page refcount OOPS:
kernel BUG at drivers/gpu/drm/i915/i915_gem.c:1444!
...
RIP: 0010:[<ffffffffa0093026>] [<ffffffffa0093026>] i915_gem_object_put_pages+0x25/0xf5 [i915]
Call Trace:
[<ffffffffa009481d>] i915_gem_object_unbind+0xc5/0x1a7 [i915]
[<ffffffffa0098ab2>] i915_gem_evict_something+0x3bd/0x409 [i915]
[<ffffffffa0027923>] ? drm_gem_object_lookup+0x27/0x57 [drm]
[<ffffffffa0093bc3>] i915_gem_object_bind_to_gtt+0x1d3/0x279 [i915]
[<ffffffffa0095b30>] i915_gem_object_pin+0xa3/0x146 [i915]
[<ffffffffa0027948>] ? drm_gem_object_lookup+0x4c/0x57 [drm]
[<ffffffffa00961bc>] i915_gem_do_execbuffer+0x50d/0xe32 [i915]
Reported-by: Shawn Starr <shawn.starr@rogers.com>
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=18902
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Use the GMBUS interface rather than direct bit banging to grab the EDID
over DDC (and for other forms of auxiliary communication with external
display controllers). The hope is that this method will be much faster
and more reliable than bit banging for fetching EDIDs from buggy monitors
or through switches, though we still preserve the bit banging as a
fallback in case GMBUS fails.
Based on an original patch by Jesse Barnes.
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
There is a second revision of B43 (a desktop gen4 part) floating around,
functionally equivalent to the original B43, so simply add the new
PCI-IDs.
Bugzilla: https://bugs.freedesktop.org/show_bugs.cgi?id=30221
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@kernel.org
With 5 places to update when adding handling for fence registers, it is
easy to overlook one or two. Correct that oversight, but fence
management should be improved before a new set of registers is added.
Bugzilla: https://bugs.freedesktop.org/show_bug?id=30199
Original patch by: Yuanhan Liu <yuanhan.liu@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@kernel.org
These are not fatal errors, so do not alarm the user by filling the
logs with *** ERROR ***. Especially as we know that g4x CRT detection
is a little sticky.
On the one hand the errors are valid since they are warning us of a
stall -- we poll the register whilst holding the mode lock so not even
the mouse will update. On the other hand, those stalls were already present
yet nobody complained.
Reported-by: Andi Kleen <andi@firstfloor.org>
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=18332
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Later initialisation of the encoder often requires that
drm_encoder_init() has already been called, for instance, initialiasing
the DDC buses.
Yet another recent regression, as 819f3fb7 depended upon these fixes
which I missed when cherry-picking.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
One problem with devices that share the DDC bus between the VGA and
DVI-I connectors is that with two devices attached we cannot know if
there is truly a monitor attached to the DVI connector. In this case, it
is preferrrable to mark the status as unknown, so that the user can
supply the known set of modes and continue to use the output.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We only need to use the analog encoder for rare devices which share the
DDC between the DVI-I and VGA connectors, so only create as needed.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The texture base address registers are in units of 256 bytes.
The original CS checker treated these offsets as bytes, so the
original check was wrong. I fixed the units in a patch during
the 2.6.36 cycle, but this ended up breaking some existing
userspace (probably due to a bug in either userspace texture allocation
or the drm texture mipmap checker). So for now, until we come
up with a better fix, just warn if the mipmap size it too large.
This will keep existing userspace working and it should be just
as safe as before when we were checking the wrong units. These
are GPU MC addresses, so if they fall outside of the VRAM or
GART apertures, they end up at the GPU default page, so this should
be safe from a security perspective.
v2: Just disable the warning. It just spams the log and there's
nothing the user can do about it.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Cc: Jerome Glisse <glisse@freedesktop.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The SDVO proxy i2c adapter wants to be able to use information stored in
the encoder, so pass that through intel_i2c rather than iterate over all
known encoders every time.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
As we currently may need to acquire a fence register during a modeset,
we need to be able to do so in an uninterruptible manner. So expose that
parameter to the callers of the fence management code.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This ensures that we do wait upon the flushes to complete if necessary
and avoid the visual tears, whilst enabling pipelined page-flips.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
There has been periodic evidence that LVDS, on at least some
panels, prefers the dividers selected by the legacy pll algo.
This patch forces the use of the legacy pll algo on RV620
LVDS panels. The old behavior (new pll algo) can be selected
by setting the new_pll module parameter to 1.
Fixes:
https://bugs.freedesktop.org/show_bug.cgi?id=30029
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Cc: stable@kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
Not 100% sure this is due to BKL removal, its most likely a combination
of that + userspace timing changes in udev/plymouth. The drm adds the sysfs
device before the driver has completed internal loading, this causes udev
to make the node and plymouth to open it before we've completed loading.
The proper solution is to delay the sysfs manipulation until later in loading
however this causes knock on issues with sysfs connector nodes, so we can use
the global mutex to serialise loading and userspace opens.
Reported-by: Toni Spets (hifi on #radeon)
Signed-off-by: Dave Airlie <airlied@redhat.com>
v2: Julien Cristau pointed out that @nondestructive results in
double-negatives and confusion when trying to interpret the parameter,
so use @force instead. Much easier to type as well. ;-)
And fix the miscompilation of vmgfx reported by Sedat Dilek.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
I pulled the wrong version of the patch from Daniel Vetter which was
missing the read barriers -- and the one that was causing all the trouble
was from i915_gem_object_put_fence_reg(), leading to GPU hangs on gen3.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
By reducing the hangcheck frequency we check less often, conserving
resources, and still detect a lock up quickly. On a fast machine with a
slow GPU (like a Core2 paired with a 945G) it is easy for the hangcheck to
misfire as we check too fast.
Also once hung and if we fail to completely reset the chip, we have a
nasty habit of proclaming a hang many times a second and generating a
strobe-like display.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Fix a regression in the previous regression fix...
In order to turn off the pipes entirely upon the first modeset, we
pretend that BIOS (or earlier module incarnation) left them active.
The first task performed by setup_initial_configuration() is to disable
all pipes and so to avoid skipping that step and so to ensure a known
configuration we need to mark all the crtcs as active.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
When separating out the prepare/commit into its own separate functions
we overlooked that the intel_crtc->dpms_mode was being used elsewhere to
check on the actual status of the pipe.
Track that bit of logic separately from the actual dpms mode, so there
is no confusion should we be able to handle multiple dpms modes, nor
any semantic conflict between prepare/commit and dpms.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This closes a couple of corner cases where we introduced and forgot
about a couple of routines that need to be called when disabling the
crtc and then re-enabling it. The code needs to be moved again so that
the common bits are shared across generations.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This commit fixes bogus CS rejection if it contains a sequence
of the following operations:
- Set the color buffer 0. track->cb[i].robj becomes non-NULL.
- Render.
- Set a larger zbuffer than the previously-set color buffer.
- Set a larger scissor area as well.
- Set the color channel mask to 0 to do depth-only rendering.
- Render. --> rejected, because track->cb[i].robj remained non-NULL,
therefore the conditional checking for the color channel mask and
friends is not performed, and the larger scissor area causes
the rejection.
This fixes bugs:
- https://bugs.freedesktop.org/show_bug.cgi?id=29762
- https://bugs.freedesktop.org/show_bug.cgi?id=28869
And maybe some others which seem to look the same.
If possible, this commit should go to stable as well.
Signed-off-by: Marek Olšák <maraeo@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
One subtest of mesa/demos/gltestperf takes 9 seconds to complete,
so to prevent an unnecessary gpu reset followed by a hardlock, I am
increasing the interval to 10 seconds after which a GPU is considered
in a locked-up state. This is on RV530. However, with a little slower GPU,
we would surpass the interval easily, so this is not a good fix
for gltestperf.
Nevertheless, this commit also fixes hardlocks in the applications which
render at speed of less than 1 frame per second, where the whole frame
consists of only one command stream. The game Tiny & Big is an example.
This bar is now lowered to 0.1 fps.
Now the question comes down to whether we should (often unsuccessfully)
reset the GPU at all? Once we have stable enough drivers, we won't have to.
Has the time come already?
If possible, this commit should go to stable as well.
Signed-off-by: Marek Olšák <maraeo@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This patch fixes rendering errors on some evergreen boards.
Hardcoding the backend map is not an optimal solution, but
a better fix is being worked on.
Similar to the fix for rv740
(6271901d82).
Fixes:
https://bugs.freedesktop.org/show_bug.cgi?id=29986
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Cc: stable@kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
Destructive load-detection is very expensive and due to failings
elsewhere can trigger system wide stalls of up to 600ms. A simple
first step to correcting this is not to invoke such an expensive
and destructive load-detection operation automatically.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=29536
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=16265
Reported-by: Bruno Prémont <bonbons@linux-vserver.org>
Tested-by: Sitsofe Wheeler <sitsofe@yahoo.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
Otherwise when disabling the output we switch to the new fb (which is
likely NULL) and skip the call to mode_set -- leaking driver private
state on the old_fb.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=29857
Reported-by: Sitsofe Wheeler <sitsofe@yahoo.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Dave Airlie <airlied@redhat.com>
Cc: stable@kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
caused by d65d65b175
need to update the radeon crtc priv native mode before using it.
Fixes:
https://bugs.freedesktop.org/show_bug.cgi?id=30049
v2: integrate v/h copy paste typo
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
I broke out my trusty i845 and found a new boot failure, which upon
inspection turned out to be a recursion within:
drm_helper_probe_single_connector_modes() -> drm_helper_hpd_irq_event()
-> intel_crt_detect() -> drm_helper_probe_single_connector_modes()
Calling drm_kms_helper_poll_enable() instead performs the desired
re-initialisation of the polling should the user have toggled the
parameter, without the recursive side-effect.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Dave Airlie <airlied@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Commit 77d07fd9d7 introduced a regression
where by not waiting for the panel to be turned off, left the panel and
PLL registers locked across the modeset. Thus the panel remaining blank.
As pointed out by Daniel Vetter, when testing LVDS it helps to open the
laptop and look at the actual panel you are purporting to test.
A second issue with the patch was that in order to modify the panel
fitter before gen5, the pipe and the panel must have be completely
powered down. So we wait.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The documentation says that an SDVO command takes a maximum of 15us to be
processed by the device, and that it is sufficient to read the status byte
3 times (whilst the command is still in the PENDING state) for the driver
to be confident that sufficient time has elapsed.
We err on the safe side and try 5 times before giving up.
The only question that remains: was the old behaviour derived by
experiments with real hardware?
A look into the murky history of UMS, implies that the behaviour was
accidental and the current retry mechanism was solely designed to catch
the status byte indicating PENDING with no reference to hardware
behaviour. (commit ac9181c014638dbeb334b40b4029d0ccb2b7a0fc in
xf86-video-intel)
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Avoid a potentially long busy-wait if we not in the process of
atomically switching to the kdb console.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We just assume that it will happen in a timely manner. A variant of this
patch was first written and tested by Arjan van de Van.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Remove our redundant udelay() as the timings are already handled by the
i2c-algo-bit controller.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The purpose is to make the code much easier to read and therefore reduce
the possibility for bugs.
A side effect is that it also makes it much easier for the compiler,
reducing the object size by 4k -- from just a few functions!
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Refactor the common code into seperate functions and use the MIN(large,
small) buffer calculation for self-refresh watermarks.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We need to track different state on each generation in order to detect
when we need to refresh the FBC registers.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Thermal reporting may not be enabled by default on some machines, so
enable the appropriate bits to allow IPS to get the data it needs from
the CPU thermal device.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
* 'drm-intel-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/ickle/drm-intel:
drm/i915: don't enable self-refresh on Ironlake
drm/i915: Double check that the wait_request is not pending before warning
Revert "drm/i915: Warn if we run out of FIFO space for a mode"
Revert "drm/i915: Allow LVDS on pipe A on gen4+"
Revert "drm/i915: Enable RC6 on Ironlake."
TU size is only part of the M1 and M2 regs, not the N regs. This keeps
us from overwriting a reserved field.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Easier to read, and will pair up with a disable function.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
eDP panels require these to be set up prior to panel power sequencing,
or they'll fail to power on due to an "asset not ready" check. And of
course, eDP panels attached to anything other than DP_A need them
enabled regardless, since they'll be driven from the CPU through FDI out
to the PCH.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This will allow us to optimize our prepare/commit paths a bit better.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
[ickle: minor tweak to handle the cursor across pipe resizing]
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This was just a workaround for some broken Ironlake CRTC code.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
So we can use it for CRTC prepare/commit.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This way we can also use it in CRTC prepare/commit. Also makes it
easier to split out FDI and other code.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
create_singlethreaded_workqueue() is being phased out for a new
concurrency managed task infrastructure.
Adapt our workqueue constructor to explicitly create a domain that only
allows the execution of a single task at any time. All the tasks are
expected to require the dev->struct_mutex, so would block concurrency of
other tasks if we allow more than a single i915 task to be run at once.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We don't know how to enable it safely, especially as outputs turn on and
off. When disabling LP1 we also need to make sure LP2 and 3 are already
disabled.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=29173
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=29082
Reported-by: Chris Lord <chris@linux.intel.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Tested-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: stable@kernel.org
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Currently we have a exact mapping of a connector onto an encoder for its
whole lifetime. Make this an explicit property of the structure and so
simplify the code.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Why iterate all the crtcs to find the pipe, when we already know which
crtc is attached to which pipe?
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
[Patch is slightly larger than is strictly necessary to fixup
surrounding checkpatch.pl errors.]
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
If we are busy, then we may have woken up the wait_request handler but
not yet serviced it before the hang check fires. So in hang check,
double check that the i915_gem_do_wait_request() is still pending the
wake-up before declaring all hope lost.
Fixes regression with e78d73b16b.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=30073
Reported-and-tested-by: Sitsofe Wheeler <sitsofe@yahoo.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Otherwise we may not be able to train the DP link.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
When turning on or off the VDD AUX bit, we need to give the panel time
to start or stop or AUX transactions may fail.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Mode set sequence outlines when the AUX VDD bit should be set and
cleared, and it's separate from the panel power sequence.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Mode set sequence requires that we start training, then enable the
panel, then complete training. So split the DP training function into
two parts; the first enables the DP port and sets training pattern 1 and
the second completes the training.
As part of this, remove some redundant function args from the various DP
handling functions and use the intel_dp fields everywhere we can.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
[ickle: removed first ironlake_edp_backlight_on() on advice of jbarnes]
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Mode setting sequence specifies that we use VDD AUX for configuration
and detection, and early in the mode set sequence. Only later (after
DP_A has started training) should we actually enable panel power.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
[ickle: checkpatch.pl complaining about whitespace]
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Fix the test so we don't try to use the 450MHz refclk on PCH attached
eDP.
References:
https://bugs.freedesktop.org/show_bug.cgi?id=29141
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
snprintf() returns the number of bytes which would have been used if
there was enough space. It can be larger than the size of the buffer.
Obviously in this case the buffer is large enough but everyone just
copy and pastes this code so it's better to limit it and set a good
example.
Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Use the detection from intel-gtt.ko instead. Hooray!
Also move the stolen mem allocator to the other gtt stuff in dev_prv->mem.
v2: Chris Wilson noted that my error handling was crap. Fix it. He also
said that this fixes a problem on his i845. Indeed, i915_probe_agp
misses a special case for i830/i845 stolen mem detection.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=25476
Cc: stable@kernel.org
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Add a few definitions to it that are already shared and that will
be shared in the future (like the number of stolen entries).
No functional changes in here.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
It seems to be possible to program a new mode without disabling the panel
if the panel fitter setup doesn't change. Add support for that.
Signed-off-by: Matthew Garrett <mjg@redhat.com>
Acked-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We really need a macro to test whether a given connector has a panel
attached rather than sprinkling HAS_PCH_SPLIT/IS_eDP/has_edp_encoder
etc all over. In the meantime, fix the bug...
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
[ickle: tidy up the duplicity in the conditionals]
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Make them match the others and add BPP definitions.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The GPU records whether it is currently waiting for a completion of a
WAIT_FOR_EVENT in the RB_WAIT bit in the ringbuffer control registers.
On third generation chipsets and later, a write of 1 to this bit breaks
the hang and returns the GPU to arbitration, i.e. the GPU should
continue executing the reminder of the batchbuffer and return to normal
operations.
By adding this to hangcheck we can avoid a full GPU reset under these
conditions.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
If we disable the pipe and the GPU is currently waiting on a scanline
WAIT_FOR_EVENT, the GPU will hang. Fortunately, there is a magic bit
which we can write on i915+ to break this wait after disabling the
pipe.
References:
Bug 29252 - [Arrandale] Hung WAIT_FOR_EVENT when running rss-glx-skyrocket
https://bugs.freedesktop.org/show_bug.cgi?id=29252
Bug 28964 - [i965gm] GPU infinite MI_WAIT_FOR_EVENT while watching video in Totem
https://bugs.freedesktop.org/show_bug.cgi?id=28964
and many others.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Hopefully this is a contributing factor to the spurious TV detection
repoted by Ivan Bulatovic and others.
References:
Bug 16871 - "TV1 connected" with no tv
https://bugzilla.kernel.org/show_bug.cgi?id=16871
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reported-by: Ivan Bulatovic <combuster@gmx.com>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
There were two instances of code to control the panel backlight and
neither handled the complete set of device variations.
Fixes:
Bug 29716 - [GM965] Regression: Backlight resets to minimum when changing resolution
https://bugs.freedesktop.org/show_bug.cgi?id=29716
And a bug on one of my PineView boxes which overflowed the backlight
value.
Incorporates part of a similar patch by Matthew Garrett that exposes a
native Intel backlight controller.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We do it whilst configuring dev->mode_config, so remove the out-of-place
earlier initialisation.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This spinlock only served debugging purposes in a time when we could not
be sure of the mutex ever being released upon a GPU hang. As we now
should be able rely on hangcheck to do the job for us (and that error
reporting should not itself require the struct mutex) we can kill the
incomplete attempt at protection.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We have no idea why we request a SyncFlush via INSTPM at that point in
time -- we certainly never check for its completion...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Alexander reported that the compilation of intel_overlay.c was failing
due to an inclusion that was only valid with CONFIG_DEBUG_FS. As the
whole error reporting is only useful with debugfs enabled, remove all
the redundant error state collection code when compiling without
CONFIG_DEBUG_FS.
Reported-by: Alexander Lam <lambchop468@gmail.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Slightly easier to follow than the state machine and now possible as the
control structure is opaque and hw_wedged is no longer interferred with.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
During DPMS we currently do not want the overlay code to be
interruptible, so pass that information down and only take the
uninterrruptible paths.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
On i830, there exists a bug where an overlay on pipe B requires the mode
clock on pipe A in order to activate. So workaround this by activating
pipe A when trying to enable the overlay on pipe B.
References:
[Bug 29007] GPU hang on video playback with overlay
https://bugs.freedesktop.org/show_bug.cgi?id=29007
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
By allocating the request prior to writing to the ringbuffer, we can
abort the operation without leaving the GPU in an inconsistent state.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Inline the call to wait_flip() and simplify the resulting code.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We can program the h/w to first wait on the flip and then switch off
without relying on s/w intervention. This removes the need for a double
step switch off, bringing much rejoicing.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The scoping of the validity of the mapping is thus clarified.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The only time where an atomic mapping is required is during
error-capture and there we cannot use the default slot, but need to
specifically use one of the IRQ slots. So separate out the two
conditions and use the atomic mapping only when appropriate.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Just makes sure that writes are not being aliased by the CPU cache and
do make it out to main memory.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=24977
Cc: stable@kernel.org
... take advantage of the new implicit request issuing of
i915_wait_request.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
One caller (for the pageflip support) wants a purely pipelined flush.
Distinguish this case by a new parameter. This will also be useful
later on for pipelined fencing.
v2: Simplify the code by depending upon the implicit request emitting
of i915_wait_request.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
[ickle: And drop the non-interruptible support in the process.]
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
By moving one i915_add_request we can solely depend on the new
auto-seqno-numbering behaviour.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
i915_gem_object_move_to_active can handle zero seqno for us now.
And not emitting a request is not fatal here - we'll try to emit
a new one if we have to wait for some rendering to complete.
In case this assumption ever gets accidentally broken, there's already
a BUG_ON to catch it in i915_do_wait_request.
So just silently ignore ENOMEM here instead of screwing up the whole
drm.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
... instead of threading flush_domains through the execbuf code to
i915_add_request.
With this change 2 small cleanups are possible (likewise the majority
of the patch):
- The flush_domains parameter of i915_add_request is always 0. Drop it
and the corresponding logic.
- Ditto for the seqno param of i915_gem_process_flushing_list.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Previously I thought that one interrupt per batchbuffer should be
enough. Now tedious benchmarking showed this to be wrong.
Therefore track whether any commands have been isssued with a future
seqno (like pipelined fencing changes or flushes). If this is the case
emit a request before issueing the batchbuffer.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Now that we can move objects to the active list without already having
emitted a request, move the flushing list handling into i915_gem_flush.
This makes more sense and allows to drop a few i915_add_request calls
that are not strictly necessary.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Sometimes (like when flushing in preparation of batchbuffer execution)
we know that we'll emit a request but haven't yet done so. Allow this
case by simply taking the next seqno by default. Ensure that a request
is eventually emitted before waiting for an request by issuing it
in i915_wait_request iff this is not yet done.
Also replace one open-coded version of i915_gem_object_wait_rendering,
to prevent future code-diversion.
Chris Wilson asked me to explain and clarify what this patch does and why.
Here it goes:
Old way of moving objects onto the active list and associating them with a
reques:
1. i915_add_request + store the returned seqno somewhere
2. i915_gem_object_move_to_active (with the stored seqno as parameter)
For the current users, this is all fine. But I'd like to associate objects
(and fence regs) with the batchbuffer request deep down in the execbuf
call-chain. I thought about three ways of implementing this.
a) Don't care, just emit request when we need a new seqno. When heavily
pipelining fence reg changes, this would have caused tons of superflous
request (and corresponding irqs).
b) Thread all changed fences, objects, whatever through the execbuf-maze,
so that when we emit a request, we can store the new seqno at all the right
places.
c) Kill that seqno-threading-around business by simply storing the next
seqno, i.e. allow 2. to be done before 1. in the above sequence.
I've decided to implement c) (in this patch). The following patches are
just fall-out that resulted from this small conceptual change.
* We can handle the flushing list processing where we actually emit a flush
(i915_gem_flush and i915_retire_commands) instead of in i915_add_request.
The code makes IMHO more sense this way (and i915_add_request looses the
flush_domains parameter, obviously).
* We can avoid emitting unnecessary requests. IMHO there's no point in
emitting more than one request per batchbuffer (with or without an
corresponding irq).
* By enforcing 2. before 1. ordering in the above sequence the seqno
argument of i915_gem_object_move_to_active is redundant and can be
dropped.
v2: Now i915_wait_request issues request if it is not yet emitted.
Also introduce i915_gem_next_request_seqno(dev) just in case we ever
need to do some prep work before using a new seqno.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
[ickle: Keep i915_gem_object_set_to_display_plane() uninterruptible.]
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Useful for capturing register read/write traces to send to the hw guys.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Instead of sleeping for an arbitrary length of time (the documentation
fails to specify how long to wait for) wait until the load detection has
changed state (or at most the 20ms as before).
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
With the extra intel_wait_for_vblank added in commit
9d0498a2bf periodic stalls were being
triggered (which were detected by i915_hangcheck_elapsed). Partially
revert this change for now.
Signed-off-by: Sitsofe Wheeler <sitsofe@yahoo.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Fix a minor confusion between intel_page_flip_finish(pipe) and
intel_page_flip_finish_plane(plane) -- should have no effect as
currently we map pipe 0 to plane 0 (and pipe 1 to plane 1).
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
My Samsung N210 has a VBT with DEVICE_TYPE_INT_LFP with a zero
addin-offset. With the check in place, the panel was declared absent.
v2: Only trust BIOS writers that have graduated to writing OpRegions.
(We are all doomed.)
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Zhao Yakui <yakui.zhao@intel.com>
Cc: Adam Jackson <ajax@redhat.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
It is recommended that we use the Video BIOS tables that were copied
into the OpRegion during POST when initialising the driver. This saves
us from having to furtle around inside the ROM ourselves and possibly
allows the vBIOS to adjust the tables prior to initialisation.
On some systems, such as the Samsung N210, there is no accessible VBIOS
and the only means of finding the VBT is through the OpRegion.
v2: Rearrange the code so that ASLE is enabled along with ACPI
v3: Enable OpRegion parsing even without ACPI
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Garrett <mjg@redhat.com>
It's part of the generic Intel driver infrastructure so rename it in
prepreparation for using it for VBT.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
If we don't flush the write then we can not be sure that the border
colour will have taken effect by the time we try to read it back.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
wait_for() uses msleep() to yield the cpu whilst spinning waiting for a
register to change. kdb asserts that mode changes are atomic and so
prohibits msleep. The alternative would be to use mdelay or to simply
probe the register more often instead of busy waiting.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Jesse's feedback from using the wait_for() macro was that the msleep
argument was that it was superfluous and made the macro more difficult
to use and to read. As the actually amount of time to sleep is not
critical, the crucial part is to sleep and let the processor schedule
something else whilst we wait for the event, replace the argument with a
hardcoded value.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
ums-gem code correctly cancels the retire work (at lastclose time),
kms does not do so. Fix this by canceling the work right after ideling
the gpu.
While staring at the code I noticed that the work function is not
static. Fix this, too.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
When the module unloads, all users should be gone, hence all bo references
held by userspace, too. This should already result in an idle ringbuffer.
Still, be paranoid and idle gem before starting the unload dance.
Also kill the call to i915_gem_lastclose under an if (kms), it's a noop
for kms.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Kill any outstanding unpin_work when destroying the corresponding
crtc. Then flush the workqueue before the gem teardown, in case
any unpin work is still outstanding.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
idle_work wasn't cleaned up at all. It takes &dev->struct_mutex, but
accesss the mode_config crtc list (without any other locking!). Hence
this work needs to be canceled before calling drm_mode_config_cleanup.
As evidenced by the kernel's object debuggin code, the current code
also cleans up the timer to early (it gets rearmed). So move it right
before the final cleanup (it seems to work).
Also unconditionally set up the idle_timer in intel_increase_pllclock.
If we're unlucky the timer might fire right away, rendering the call
in the modesetting teardown pointless.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
With kms, interrupts now get disabled in the modesetting cleanup. So
free the error state afterwards, it currently gets allocated in
the interrupt handler.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
