reset_counter will be incremented twice per successful
reset. Odd values mean reset is in progress and even values
mean that reset has completed.
Reset status ioctl introduced in following commit
needs to deliver global reset count to userspace so
use reset_counter to derive the actual reset count
for the gpu
Note that reset in progress is enough to increment
the counter.
v2: wedged equals reset in progress (Daniel Vetter)
v3: Fixed stale comments (Damien Lespiau)
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Make the cur_delay limiting code a bit less prone to typo errors
by using clamp_t().
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>
When the hardware frame counter reads 0xffffff and we're already past
vblank start, we'd return 0x1000000 as the vblank counter value. Once
we'd cross into the next frame's active portion, the vblank counter
would wrap to 0. So we're reporting two different vblank counter values
for the same frame.
Fix the problem by masking the cooked value by 0xffffff to make sure
the counter wraps already after vblank start.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
- Use a for_each_loop and add the corresponding #defines.
- Drop the _ILK postfix on the existing DE_PIPE_VBLANK macro for
consistency with everything else.
- Also use macros (and add the missing one for plane flips) for the
ivb display interrupt handler.
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
[danvet: Drop the useless parens that Ville spotted.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Request by Ville in his review of the CRC stuff. This converts
everything but ilk_display_irq_handler since that needs a bit more
than a simple search&replace to look nice.
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Otherwise QA will report this as a real hang when running igt
ZZ_missed_irq.
v2: Actually test the right stuff and really shut up the DRM_ERROR
output ...
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=70747
Tested-by: lu hua <huax.lu@intel.com>
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Daniel pointed out that it was hard to get anything lockless to work
correctly, so don't even try for this non critical piece of code and
just use a spin lock.
v2: Make intel_pipe_crc->opened a bool
v3: Use assert_spin_locked() instead of a comment (Daniel Vetter)
v4: Use spin_lock_irq() in the debugfs functions (they can only be
called from process context),
Use spin_lock() in the pipe_crc_update() function that can only be
called from an interrupt handler,
Use wait_event_interruptible_lock_irq() when waiting for data in the
cicular buffer to ensure proper locking around the condition we are
waiting for. (Daniel Vetter)
Suggested-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
- Give them an _irq_handler postfix, like all the other irq stuff.
- Shuffle the DEBUG_FS=n dummy functions around a bit. This is prep
work to extract all the crc debug stuff into intel_display_testing.c
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
And throw in a tiny for_each_pipe refactoring for gen2.
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Should work down to gen2. The #defines for the interrupt sources are
already there in PIPESTAT and are the same on all gmch platforms for
gen2 up to vlv.
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
hw designers decided to change the CRC registers and coalesce them all
into one. Otherwise nothing changed. I've opted for a new hsw_ version
to grab the crc sample since hsw+1 will have the same crc registers,
but different interrupt source registers. So this little helper
function will come handy there.
Also refactor the display error handler with a neat pipe loop.
v2: Use for_each_pipe.
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Suggested by Ville.
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We enable the interrupt unconditionally and only control it
through the enable bit in the CRC control register.
v2: Extract per-platform helpers to compute the register values.
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The ringbuffer update logic should always be the same, but different
platforms have different amounts of CRC registers. Hence extract it.
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Also use #ifdef to keep consistent with all other such cases.
Cc: Damien Lespiau <damien.lespiau@intel.com>
Acked-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
seq_file is not quite the right interface for these ones. We have a
circular buffer with a new entry per vblank on one side and a process
wanting to dequeue the CRC with a read().
It's quite racy to wait for vblank in user land and then try to read a
pipe_crc file, sometimes the CRC interrupt hasn't been fired and we end
up with an EOF.
So, let's have the read on the pipe_crc file block until the interrupt
gives us a new entry. At that point we can wake the reading process.
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
This shouldn't happen as the buffer is freed after disable pipe CRCs,
but better be safe than sorry.
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
There are a few good properties to a circular buffer, for instance it
has a number of entries (before we were always dumping the full buffer).
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
There are several points in the display pipeline where CRCs can be
computed on the bits flowing there. For instance, it's usually possible
to compute the CRCs of the primary plane, the sprite plane or the CRCs
of the bits after the panel fitter (collectively called pipe CRCs).
v2: Quite a bit of rework here and there (Damien)
Signed-off-by: Shuang He <shuang.he@intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
[danvet: Fix intermediate compile file reported by Wu Fengguang's
kernel builder.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Gen2 doesn't have a hardware frame counter that can be read out. Just
provide a stub .get_vblank_counter() that always returns 0 instead of
trying to read non-existing registers.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Gen2 doesn't have the pixelcount register that gen3 and gen4 have.
Instead we must use the scanline counter like we do for ctg+.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The DSL register increments at the start of horizontal sync, so it
manages to miss the entire active portion of the current line.
Improve the get_scanoutpos accuracy a bit when the scanout position is
close to the start or end of vblank. We can do that by double checking
the DSL value against the vblank status bit from ISR.
Cc: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: mario.kleiner.de@gmail.com
Tested-by: mario.kleiner.de@gmail.com
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The reported scanout position must be relative to the end of vblank.
Currently we manage to fumble that in a few ways.
First we don't consider the case when vtotal != vbl_end. While that
isn't very common (happens maybe only w/ old panel fitting hardware),
we can fix it easily enough.
The second issue is that on pre-CTG hardware we convert the pixel count
to horizontal/vertical components at the very beginning, and then forget
to adjust the horizontal component to be relative to vbl_end. So instead
we should keep our numbers in the pixel count domain while we're
adjusting the position to be relative to vbl_end. Then when we do the
conversion in the end, both vertical _and_ horizontal components will
come out correct.
v2: Change position to int from u32 to avoid sign issues
Cc: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: mario.kleiner.de@gmail.com
Tested-by: mario.kleiner.de@gmail.com
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We have all the information we need in the mode structure, so going and
reading it from the hardware is pointless, and slower.
We never populated ->get_vblank_timestamp() in the UMS case, and as that
is the only way we'd ever call ->get_scanout_position(), we can
completely ignore UMS in i915_get_crtc_scanoutpos().
Also reorganize intel_irq_init() a bit to clarify the KMS vs. UMS
situation.
v2: Drop UMS code
Cc: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: mario.kleiner.de@gmail.com
Tested-by: mario.kleiner.de@gmail.com
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The old style frame counter increments at the start of active video.
However for i915_get_vblank_counter() we want a counter that increments
at the start of vblank.
Fortunately the low frame counter register also contains the pixel
counter for the current frame. We can can compare that against the
vblank start pixel count to determine if we need to increment the
frame counter by 1 to get the correct answer.
Also reorganize the function pointer assignments in intel_irq_init() a
bit to avoid confusing people.
Cc: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We lost the ability to capture the first error for a stuck ring in the
recent hangcheck robustification. Whilst both error states are
interesting (why does the GPU not recover is also essential to debug),
our primary goal is to fix the initial hang and so we need to capture
the first error state upon taking hangcheck action.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
After applying wait-boost we often find ourselves stuck at higher clocks
than required. The current threshold value requires the GPU to be
continuously and completely idle for 313ms before it is dropped by one
bin. Conversely, we require the GPU to be busy for an average of 90% over
a 84ms period before we upclock. So the current thresholds almost never
downclock the GPU, and respond very slowly to sudden demands for more
power. It is easy to observe that we currently lock into the wrong bin
and both underperform in benchmarks and consume more power than optimal
(just by repeating the task and measuring the different results).
An alternative approach, as discussed in the bspec, is to use a
continuous threshold for upclocking, and an average value for downclocking.
This is good for quickly detecting and reacting to state changes within a
frame, however it fails with the common throttling method of waiting
upon the outstanding frame - at least it is difficult to choose a
threshold that works well at 15,000fps and at 60fps. So continue to use
average busy/idle loads to determine frequency change.
v2: Use 3 power zones to keep frequencies low in steady-state mostly
idle (e.g. scrolling, interactive 2D drawing), and frequencies high
for demanding games. In between those end-states, we use a
fast-reclocking algorithm to converge more quickly on the desired bin.
v3: Bug fixes - make sure we reset adj after switching power zones.
v4: Tune - drop the continuous busy thresholds as it prevents us from
choosing the right frequency for glxgears style swap benchmarks. Instead
the goal is to be able to find the right clocks irrespective of the
wait-boost.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Kenneth Graunke <kenneth@whitecape.org>
Cc: Stéphane Marchesin <stephane.marchesin@gmail.com>
Cc: Owen Taylor <otaylor@redhat.com>
Cc: "Meng, Mengmeng" <mengmeng.meng@intel.com>
Cc: "Zhuang, Lena" <lena.zhuang@intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
If we encounter a situation where the CPU blocks waiting for results
from the GPU, give the GPU a kick to boost its the frequency.
This should work to reduce user interface stalls and to quickly promote
mesa to high frequencies - but the cost is that our requested frequency
stalls high (as we do not idle for long enough before rc6 to start
reducing frequencies, nor are we aggressive at down clocking an
underused GPU). However, this should be mitigated by rc6 itself powering
off the GPU when idle, and that energy use is dependent upon the workload
of the GPU in addition to its frequency (e.g. the math or sampler
functions only consume power when used). Still, this is likely to
adversely affect light workloads.
In particular, this nearly eliminates the highly noticeable wake-up lag
in animations from idle. For example, expose or workspace transitions.
(However, given the situation where we fail to downclock, our requested
frequency is almost always the maximum, except for Baytrail where we
manually downclock upon idling. This often masks the latency of
upclocking after being idle, so animations are typically smooth - at the
cost of increased power consumption.)
Stéphane raised the concern that this will punish good applications and
reward bad applications - but due to the nature of how mesa performs its
client throttling, I believe all mesa applications will be roughly
equally affected. To address this concern, and to prevent applications
like compositors from permanently boosting the RPS state, we ratelimit the
frequency of the wait-boosts each client recieves.
Unfortunately, this techinique is ineffective with Ironlake - which also
has dynamic render power states and suffers just as dramatically. For
Ironlake, the thermal/power headroom is shared with the CPU through
Intelligent Power Sharing and the intel-ips module. This leaves us with
no GPU boost frequencies available when coming out of idle, and due to
hardware limitations we cannot change the arbitration between the CPU and
GPU quickly enough to be effective.
v2: Limit each client to receiving a single boost for each active period.
Tested by QA to only marginally increase power, and to demonstrably
increase throughput in games. No latency measurements yet.
v3: Cater for front-buffer rendering with manual throttling.
v4: Tidy up.
v5: Sadly the compositor needs frequent boosts as it may never idle, but
due to its picking mechanism (using ReadPixels) may require frequent
waits. Those waits, along with the waits for the vrefresh swap, conspire
to keep the GPU at low frequencies despite the interactive latency. To
overcome this we ditch the one-boost-per-active-period and just ratelimit
the number of wait-boosts each client can receive.
Reported-and-tested-by: Paul Neumann <paul104x@yahoo.de>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=68716
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Kenneth Graunke <kenneth@whitecape.org>
Cc: Stéphane Marchesin <stephane.marchesin@gmail.com>
Cc: Owen Taylor <otaylor@redhat.com>
Cc: "Meng, Mengmeng" <mengmeng.meng@intel.com>
Cc: "Zhuang, Lena" <lena.zhuang@intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
[danvet: No extern for function prototypes in headers.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
When we switched to always using a timeout in conjunction with
wait_seqno, we lost the ability to detect missed interrupts. Since, we
have had issues with interrupts on a number of generations, and they are
required to be delivered in a timely fashion for a smooth UX, it is
important that we do log errors found in the wild and prevent the
display stalling for upwards of 1s every time the seqno interrupt is
missed.
Rather than continue to fix up the timeouts to work around the interface
impedence in wait_event_*(), open code the combination of
wait_event[_interruptible][_timeout], and use the exposed timer to
poll for seqno should we detect a lost interrupt.
v2: In order to satisfy the debug requirement of logging missed
interrupts with the real world requirments of making machines work even
if interrupts are hosed, we revert to polling after detecting a missed
interrupt.
v3: Throw in a debugfs interface to simulate broken hw not reporting
interrupts.
v4: s/EGAIN/EAGAIN/ (Imre)
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Imre Deak <imre.deak@intel.com>
[danvet: Don't use the struct typedef in new code.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We only wish to know the value of seqno when emitting the tracepoint, so
move the query from a parameter to the macro to inside the conditional
macro body so that the query is only evaluated when required.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.14 (GNU/Linux)
iQEcBAABAgAGBQJSQMORAAoJEHm+PkMAQRiGj14H/1bjhtfNjPdX7MVQAzA+WpwX
s7h1IQu2Si9S5S1lBiM2sBTOssVcmfheO9x4yqm7JNOD1RnssWKOM3q+zVOLstwd
GD3gluJPeraD5EyYSqEJ9ILPQ3gbxb4wOlT0Z291TW6E8XhLRr0RTOJPksRsgvLH
Ckm9uJh6ArS6ZXfXiaDQfd+xHAQJkUfW6nMSA0g9ZO9C6KIDRvcbUmrY3m4HhfIk
mK0TXCBs+AXGDIjTEB8JgIQL/5y1Qn0c4R+2uTU/4YWwyLvJTV1e44kGoleukMMT
6Pw/TNlUEN161dbSaqCyF3sfXHDYQ5valycI2PDgitMtPSxbzsU1VDizS8+daRg=
=lEmF
-----END PGP SIGNATURE-----
Merge tag 'v3.12-rc2' into drm-intel-next
Backmerge Linux 3.12-rc2 to prep for a bunch of -next patches:
- Header cleanup in intel_drv.h, both changed in -fixes and my current
-next pile.
- Cursor handling cleanup for -next which depends upon the cursor
handling fix merged into -rc2.
All just trivial conflicts of the "changed adjacent lines" type:
drivers/gpu/drm/i915/i915_gem.c
drivers/gpu/drm/i915/intel_display.c
drivers/gpu/drm/i915/intel_drv.h
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We currently disable the ERR_INT interrupts while running the IRQ
handler because we fear that if we do an unclaimed register access
from inside the IRQ handler we'll keep triggering the IRQ handler
forever.
The problem is that since we always disable the ERR_INT interrupts at
the IRQ handler, when we get a FIFO underrun we'll always print both
messages:
- "uncleared fifo underrun on pipe A"
- "Pipe A FIFO underrun"
Because the "was_enabled" variable from
ivybridge_set_fifo_underrun_reporting will always be false (since we
disable ERR int at the IRQ handler!).
Instead of actually fixing ivybridge_set_fifo_underrun_reporting,
let's just remove the "disable ERR_INT during the IRQ handler" code.
As far as we know we shouldn't really be triggering ERR_INT interrupts
from the IRQ handler, so if we ever get stuck in the endless loop of
interrupts we can git-bisect and revert (and we can even bisect and
revert this patch in case I'm just wrong). As a bonus, our IRQ handler
is now simpler and a few nanoseconds faster.
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We'd only ever used this define to denote whether or not we have the
dynamic parity feature (DPF) and never to determine whether or not L3
exists. Baytrail is a good example of where L3 exists, and not DPF.
This patch provides clarify in the code for future use cases which might
want to actually query whether or not L3 exists.
v2: Add /* DPF == dynamic parity feature */
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Certain HSW SKUs have a second bank of L3. This L3 remapping has a
separate register set, and interrupt from the first "slice". A slice is
simply a term to define some subset of the GPU's l3 cache. This patch
implements both the interrupt handler, and ability to communicate with
userspace about this second slice.
v2: Remove redundant check about non-existent slice.
Change warning about interrupts of unknown slices to WARN_ON_ONCE
Handle the case where we get 2 slice interrupts concurrently, and switch
the tracking of interrupts to be non-destructive (all Ville)
Don't enable/mask the second slice parity interrupt for ivb/vlv (even
though all docs I can find claim it's rsvd) (Ville + Bryan)
Keep BYT excluded from L3 parity
v3: Fix the slice = ffs to be decremented by one (found by Ville). When
I initially did my testing on the series, I was using 1-based slice
counting, so this code was correct. Not sure why my simpler tests that
I've been running since then didn't pick it up sooner.
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
This reduces dmesg noise when there's a glitch on the hpd line, or there
are more than one connectors on the same hpd line and only one of them
changes.
While at it, switch to use the friendly status names instead of numbers.
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
My g33 here seems to be shockingly good at hitting them all. This time
around kms_flip/flip-vs-panning-vs-hang blows up:
intel_crtc_wait_for_pending_flips correctly checks for gpu hangs and
if a gpu hang is pending aborts the wait for outstanding flips so that
the setcrtc call will succeed and release the crtc mutex. And the gpu
hang handler needs that lock in intel_display_handle_reset to be able
to complete outstanding flips.
The problem is that we can race in two ways:
- Waiters on the dev_priv->pending_flip_queue aren't woken up after
we've the reset as pending, but before we actually start the reset
work. This means that the waiter doesn't notice the pending reset
and hence will keep on hogging the locks.
Like with dev->struct_mutex and the ring->irq_queue wait queues we
there need to wake up everyone that potentially holds a lock which
the reset handler needs.
- intel_display_handle_reset was called _after_ we've already
signalled the completion of the reset work. Which means a waiter
could sneak in, grab the lock and never release it (since the
pageflips won't ever get released).
Similar to resetting the gem state all the reset work must complete
before we update the reset counter. Contrary to the gem reset we
don't need to have a second explicit wake up call since that will
have happened already when completing the pageflips. We also don't
have any issues that the completion happens while the reset state is
still pending - wait_for_pending_flips is only there to ensure we
display the right frame. After a gpu hang&reset events such
guarantees are out the window anyway. This is in contrast to the gem
code where too-early wake-up would result in unnecessary restarting
of ioctls.
Also, since we've gotten these various deadlocks and ordering
constraints wrong so often throw copious amounts of comments at the
code.
This deadlock regression has been introduced in the commit which added
the pageflip reset logic to the gpu hang work:
commit 96a02917a0
Author: Ville Syrjälä <ville.syrjala@linux.intel.com>
Date: Mon Feb 18 19:08:49 2013 +0200
drm/i915: Finish page flips and update primary planes after a GPU reset
v2:
- Add comments to explain how the wake_up serves as memory barriers
for the atomic_t reset counter.
- Improve the comments a bit as suggested by Chris Wilson.
- Extract the wake_up calls before/after the reset into a little
i915_error_wake_up and unconditionally wake up the
pending_flip_queue waiters, again as suggested by Chris Wilson.
v3: Throw copious amounts of comments at i915_error_wake_up as
suggested by Chris Wilson.
Cc: stable@vger.kernel.org
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Score and action reveals what all the rings were doing
and why hang was declared. Add idle state so that
we can distinguish between waiting and idle ring.
v2: - add idle as a hangcheck action
- consensed hangcheck status to single line (Chris)
- mark active explicitly when we are making progress (Chris)
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Since we've started to clean up pending flips when the gpu hangs in
commit 96a02917a0
Author: Ville Syrjälä <ville.syrjala@linux.intel.com>
Date: Mon Feb 18 19:08:49 2013 +0200
drm/i915: Finish page flips and update primary planes after a GPU reset
the gpu reset work now also grabs modeset locks. But since work items
on our private work queue are not allowed to do that due to the
flush_workqueue from the pageflip code this results in a neat
deadlock:
INFO: task kms_flip:14676 blocked for more than 120 seconds.
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
kms_flip D ffff88019283a5c0 0 14676 13344 0x00000004
ffff88018e62dbf8 0000000000000046 ffff88013bdb12e0 ffff88018e62dfd8
ffff88018e62dfd8 00000000001d3b00 ffff88019283a5c0 ffff88018ec21000
ffff88018f693f00 ffff88018eece000 ffff88018e62dd60 ffff88018eece898
Call Trace:
[<ffffffff8138ee7b>] schedule+0x60/0x62
[<ffffffffa046c0dd>] intel_crtc_wait_for_pending_flips+0xb2/0x114 [i915]
[<ffffffff81050ff4>] ? finish_wait+0x60/0x60
[<ffffffffa0478041>] intel_crtc_set_config+0x7f3/0x81e [i915]
[<ffffffffa031780a>] drm_mode_set_config_internal+0x4f/0xc6 [drm]
[<ffffffffa0319cf3>] drm_mode_setcrtc+0x44d/0x4f9 [drm]
[<ffffffff810e44da>] ? might_fault+0x38/0x86
[<ffffffffa030d51f>] drm_ioctl+0x2f9/0x447 [drm]
[<ffffffff8107a722>] ? trace_hardirqs_off+0xd/0xf
[<ffffffffa03198a6>] ? drm_mode_setplane+0x343/0x343 [drm]
[<ffffffff8112222f>] ? mntput_no_expire+0x3e/0x13d
[<ffffffff81117f33>] vfs_ioctl+0x18/0x34
[<ffffffff81118776>] do_vfs_ioctl+0x396/0x454
[<ffffffff81396b37>] ? sysret_check+0x1b/0x56
[<ffffffff81118886>] SyS_ioctl+0x52/0x7d
[<ffffffff81396b12>] system_call_fastpath+0x16/0x1b
2 locks held by kms_flip/14676:
#0: (&dev->mode_config.mutex){+.+.+.}, at: [<ffffffffa0316545>] drm_modeset_lock_all+0x22/0x59 [drm]
#1: (&crtc->mutex){+.+.+.}, at: [<ffffffffa031656b>] drm_modeset_lock_all+0x48/0x59 [drm]
INFO: task kworker/u8:4:175 blocked for more than 120 seconds.
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
kworker/u8:4 D ffff88018de9a5c0 0 175 2 0x00000000
Workqueue: i915 i915_error_work_func [i915]
ffff88018e37dc30 0000000000000046 ffff8801938ab8a0 ffff88018e37dfd8
ffff88018e37dfd8 00000000001d3b00 ffff88018de9a5c0 ffff88018ec21018
0000000000000246 ffff88018e37dca0 000000005a865a86 ffff88018de9a5c0
Call Trace:
[<ffffffff8138ee7b>] schedule+0x60/0x62
[<ffffffff8138f23d>] schedule_preempt_disabled+0x9/0xb
[<ffffffff8138d0cd>] mutex_lock_nested+0x205/0x3b1
[<ffffffffa0477094>] ? intel_display_handle_reset+0x7e/0xbd [i915]
[<ffffffffa0477094>] ? intel_display_handle_reset+0x7e/0xbd [i915]
[<ffffffffa0477094>] intel_display_handle_reset+0x7e/0xbd [i915]
[<ffffffffa044e0a2>] i915_error_work_func+0x128/0x147 [i915]
[<ffffffff8104a89a>] process_one_work+0x1d4/0x35a
[<ffffffff8104a821>] ? process_one_work+0x15b/0x35a
[<ffffffff8104b4a5>] worker_thread+0x144/0x1f0
[<ffffffff8104b361>] ? rescuer_thread+0x275/0x275
[<ffffffff8105076d>] kthread+0xac/0xb4
[<ffffffff81059d30>] ? finish_task_switch+0x3b/0xc0
[<ffffffff810506c1>] ? __kthread_parkme+0x60/0x60
[<ffffffff81396a6c>] ret_from_fork+0x7c/0xb0
[<ffffffff810506c1>] ? __kthread_parkme+0x60/0x60
3 locks held by kworker/u8:4/175:
#0: (i915){.+.+.+}, at: [<ffffffff8104a821>] process_one_work+0x15b/0x35a
#1: ((&dev_priv->gpu_error.work)){+.+.+.}, at: [<ffffffff8104a821>] process_one_work+0x15b/0x35a
#2: (&crtc->mutex){+.+.+.}, at: [<ffffffffa0477094>] intel_display_handle_reset+0x7e/0xbd [i915]
This blew up while running kms_flip/flip-vs-panning-vs-hang-interruptible
on one of my older machines.
Unfortunately (despite the proper lockdep annotations for
flush_workqueue) lockdep still doesn't detect this correctly, so we
need to rely on chance to discover these bugs.
Apply the usual bugfix and schedule the reset work on the system
workqueue to keep our own driver workqueue free of any modeset lock
grabbing.
Note that this is not a terribly serious regression since before the
offending commit we'd simply have stalled userspace forever due to
failing to abort all outstanding pageflips.
v2: Add a comment as requested by Chris.
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Historically we've run our own driver hotplug handling in our own
work-queue, which then launched the drm core hotplug handling in the
system workqueue. This is important since we flush our own driver
workqueue in the pageflip code while hodling modeset locks, and only
the drm hotplug code grabbed these locks. But with
commit 69787f7da6
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date: Tue Oct 23 18:23:34 2012 +0000
drm: run the hpd irq event code directly
this was changed and now we could deadlock in our flip handler if
there's a hotplug work blocking the progress of the crucial unpin
works. So this broke the careful deadlock avoidance implemented in
commit b4a98e57fc
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date: Thu Nov 1 09:26:26 2012 +0000
drm/i915: Flush outstanding unpin tasks before pageflipping
Since the rule thus far has been that work items on our own workqueue
may never grab modeset locks simply restore that rule again.
v2: Add a comment to the declaration of dev_priv->wq to warn readers
about the tricky implications of using it. Suggested by Chris Wilson.
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Stuart Abercrombie <sabercrombie@chromium.org>
Reported-by: Stuart Abercrombie <sabercrombie@chromium.org>
References: http://permalink.gmane.org/gmane.comp.freedesktop.xorg.drivers.intel/26239
Cc: stable@vger.kernel.org
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Squash in a comment at the place where we schedule the work.
Requested after-the-fact by Chris on irc since the hpd work isn't the
only place we botch this.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We already have a big splashing *ERROR* for all the relevant cases of
hangs, so this one here is redudant. And it results in an unclean
dmesg when running with simulated hangs. Regression has been
introduced in
commit 05407ff889
Author: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Date: Thu May 30 09:04:29 2013 +0300
drm/i915: detect hang using per ring hangcheck_score
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=68641
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
This patch allows PC8+ states on Haswell. These states can only be
reached when all the display outputs are disabled, and they allow some
more power savings.
The fact that the graphics device is allowing PC8+ doesn't mean that
the machine will actually enter PC8+: all the other devices also need
to allow PC8+.
For now this option is disabled by default. You need i915.allow_pc8=1
if you want it.
This patch adds a big comment inside i915_drv.h explaining how it
works and how it tracks things. Read it.
v2: (this is not really v2, many previous versions were already sent,
but they had different names)
- Use the new functions to enable/disable GTIMR and GEN6_PMIMR
- Rename almost all variables and functions to names suggested by
Chris
- More WARNs on the IRQ handling code
- Also disable PC8 when there's GPU work to do (thanks to Ben for
the help on this), so apps can run caster
- Enable PC8 on a delayed work function that is delayed for 5
seconds. This makes sure we only enable PC8+ if we're really
idle
- Make sure we're not in PC8+ when suspending
v3: - WARN if IRQs are disabled on __wait_seqno
- Replace some DRM_ERRORs with WARNs
- Fix calls to restore GT and PM interrupts
- Use intel_mark_busy instead of intel_ring_advance to disable PC8
v4: - Use the force_wake, Luke!
v5: - Remove the "IIR is not zero" WARNs
- Move the force_wake chunk to its own patch
- Only restore what's missing from RC6, not everything
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Because hsw_pm_irq_handler does exactly what gen6_rps_irq_handler does
and also processes the 2 additional VEBOX bits. So merge those
functions and wrap the VEBOX bits on a HAS_VEBOX check. This
check isn't really necessary since the bits are reserved on
SNB/IVB/VLV, but it's a good documentation on who uses them.
v2: - Change IS_HASWELL check to HAS_VEBOX
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
It seems we've been doing this ever since we started processing the
RPS events on a work queue, on commit "drm/i915: move gen6 rps
handling to workqueue", 4912d04193.
The problem is: when we add work to the queue, instead of just masking
the bits we queued and leaving all the others on their current state,
we mask the bits we queued and unmask all the others. This basically
means we'll be unmasking a bunch of interrupts we're not going to
process. And if you look at gen6_pm_rps_work, we unmask back only
GEN6_PM_RPS_EVENTS, which means the bits we unmasked when adding work
to the queue will remain unmasked after we process the queue.
Notice that even though we unmask those unrelated interrupts, we never
enable them on IER, so they don't fire our interrupt handler, they
just stay there on IIR waiting to be cleared when something else
triggers the interrupt handler.
So this patch does what seems to make more sense: mask only the bits
we add to the queue, without unmasking anything else, and so we'll
unmask them after we process the queue.
As a side effect we also have to remove that WARN, because it is not
only making sure we don't mask useful interrupts, it is also making
sure we do unmask useless interrupts! That piece of code should not be
responsible for knowing which bits should be unmasked, so just don't
assert anything, and trust that snb_disable_pm_irq should be doing the
right thing.
With i915.enable_pc8=1 I was getting ocasional "GEN6_PMIIR is not 0"
error messages due to the fact that we unmask those unrelated
interrupts but don't enable them.
Note: if bugs start bisecting to this patch, then it probably means
someone was relying on the fact that we unmask everything by accident,
then we should fix gen5_gt_irq_postinstall or whoever needs the
accidentally unmasked interrupts. Or maybe I was just wrong and we
need to revert this patch :)
Note: This started to be a more real issue with the addition of the
VEBOX support since now we do enable more than just the minimal set of
RPS interrupts in the IER register. Which means after the first rps
interrupt has happened we will never mask the VEBOX user interrupts
again and so will blow through cpu time needlessly when running video
workloads.
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
[danvet: Add note that this started to matter with VEBOX much more.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
On SNB/IVB/VLV we only call gen6_rps_irq_handler if one of the IIR
bits set is part of GEN6_PM_RPS_EVENTS, but at gen6_rps_irq_handler we
add all the enabled IIR bits to the work queue, not only the ones that
are part of GEN6_PM_RPS_EVENTS. But then gen6_pm_rps_work only
processes GEN6_PM_RPS_EVENTS, so it's useless to add anything that's
not GEN6_PM_RPS_EVENTS to the work queue.
As a bonus, gen6_rps_irq_handler looks more similar to
hsw_pm_irq_handler, so we may be able to merge them in the future.
v2: - Add a WARN in case we queued something we're not going to
process.
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net> (v1)
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
If the error interrupts are already disabled, don't disable and
reenable them. This is going to be needed when we're in PC8+, where
all the interrupts are disabled so we won't risk re-enabling
DE_ERR_INT_IVB.
v2: Use dev_priv->irq_mask (Chris)
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Just like irq_mask and gt_irq_mask, use it to track the status of
GEN6_PMIMR so we don't need to read it again every time we call
snb_update_pm_irq.
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
I did some brief tests and the "new_val = pmimr" condition usually
happens a few times after exiting games.
Note: This is also prep work to track the GEN6_PMIMR register state in
dev_priv->pm_imr. This happens in the next patch.
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
[danvet: Add note to explain why we want this, as per the discussion
between Chris and Paulo.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Just like we're doing with the other IMR changes.
One of the functional changes is that not every caller was doing the
POSTING_READ.
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Mika Kuoppala