The bulk of the change is to convert the growing list of rings into an
array so that the relationship between the rings and the semaphore sync
registers can be easily computed.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Magic numbers from the specs. This is supposed to allow the PLL some
variance to improve jitter performance and VCO headroom across
manufacturing and environmental variations.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
... it's because setting the Pixel Multiply bits only takes effect once
the PLL is enabled and stable.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The pipe is always set to 8BPC, but here we were leaving whatever
previous bits were set by the BIOS in place.
Signed-off-by: Eric Anholt <eric@anholt.net>
Tested-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@kernel.org
With this change, every batchbuffer can use all available fences (save
pinned and scanout, of course) without ever stalling the gpu!
In theory. Currently the actual pipelined update of the register is
disabled due to some stability issues. However, just the deferred update
is a significant win.
Based on a series of patches by Daniel Vetter.
The premise is that before every access to a buffer through the GTT we
have to declare whether we need a register or not. If the access is by
the GPU, a pipelined update to the register is made via the ringbuffer,
and we track the last seqno of the batches that access it. If by the
CPU we wait for the last GPU access and update the register (either
to clear or to set it for the current buffer).
One advantage of being able to pipeline changes is that we can defer the
actual updating of the fence register until we first need to access the
object through the GTT, i.e. we can eliminate the stall on set_tiling.
This is important as the userspace bo cache does not track the tiling
status of active buffers which generate frequent stalls on gen3 when
enabling tiling for an already bound buffer.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
... otherwise the panel-fitter may be left enabled with random settings
and cause unintended filtering (i.e. blurring of native modes on external
panels).
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=31942
Reported-and-tested-by: Ben Kohler <bkohler@gmail.com>
Tested-by: Ciprian Docan <docan@eden.rutgers.edu>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
When trying to diagnose mysterious errors on resume, capture the
display register contents as well.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
An old and oft reported bug, is that of the GPU hanging on a
MI_WAIT_FOR_EVENT following a mode switch. The cause is that the GPU is
waiting on a scanline counter on an inactive pipe, and so waits for a
very long time until eventually the user reboots his machine.
We can prevent this either by moving the WAIT into the kernel and
thereby incurring considerable cost on every swapbuffers, or by waiting
for the GPU to retire the last batch that accesses the framebuffer
before installing a new one. As mode switches are much rarer than swap
buffers, this looks like an easy choice.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=28964
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=29252
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@kernel.org
We only ever used the PRB0, neglecting the secondary ring buffers, and
now with the advent of multiple engines with separate ring buffers we
need to excise the anachronisms from our code (and be explicit about
which ring we mean where). This is doubly important in light of the
FORCEWAKE required to read ring buffer registers on SandyBridge.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We use i915_gem_object_get_fence_reg() to do LRU tracking of the fence
registers, so stop trying to be too clever when pinning the fb->obj.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This is not known to fix any particular bugs we have, but the spec
says to do it, and the BIOS hadn't already set it up on my system.
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
a00b10c360 "Only enforce fence limits inside the GTT" also
added a fenceable/mappable disdinction when binding/pinning buffers.
This only complicates the code with no pratical gain:
- In execbuffer this matters on for g33/pineview, as this is the only
chip that needs fences and has an unmappable gtt area. But fences
are only possible in the mappable part of the gtt, so need_fence
implies need_mappable. And need_mappable is only set independantly
with relocations which implies (for sane userspace) that the buffer
is untiled.
- The overlay code is only really used on i8xx, which doesn't have
unmappable gtt. And it doesn't support tiled buffers, currently.
- For all other buffers it's a bug to pass in a tiled bo.
In short, this disdinction doesn't have any practical gain.
I've also reverted mapping the overlay and context pages as possibly
unmappable. It's not worth being overtly clever here, all the big
gains from unmappable are for execbuf bos.
Also add a comment for a clever optimization that confused me
while reading the original patch by Chris Wilson.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
In a00b10c360 "Only enforce fence limits inside the GTT"
Chris Wilson implemented an optimization to only pin framebuffers
as mappable for crtc_set_base (but not for pageflips). This breaks
the abi, eg: A double buffering mesa client might leave the last
framebuffer in unmappable space on close. A subsequent glReadPix
by a frontbuffer rendering client then goes boom. My pretty anal
mappable/unmappable consistency checking detected this, see
https://bugs.freedesktop.org/show_bug.cgi?id=31286
Chris Wilson tried to fix this in 085ce26437 by pinning
tiled framebuffers into mappable space. This
a) renders the original optimization of not forcing framebuffers
for pageflipping clients into mappable pointless because all our
scanout buffers are tiled by default.
b) doesn't solve the problem for untiled framebuffers.
So kill this. Emperically it's no gain anyway because framebuffers are
being reused by the ddx and hence there's no chance for them to get
constanly bounced between mappable and unmappable.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We should enable FDI normal training on Sandybridge/CPT system
as well.
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
[ickle: removed unrelated chunks]
Cc: stable@kernel.org
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
When merging Daniel's full-gtt patches I had a set of tweaks which I
thought I had undone. I was half right...
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=31286
Reported-by: jinjin.wang@intel.com
Reported-by: Alexey Fisher <bug-track@fisher-privat.net>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Immediate merge to resolve conflicts from applying a stability fix to
both branches.
Conflicts:
drivers/gpu/drm/i915/intel_ringbuffer.c
drivers/gpu/drm/i915/intel_ringbuffer.h
Part of the issue here was that Eric slipped in a debug hack for
testing the i915 IPS code before the intel_ips.c driver had landed.
This caused the driver to always use the full range of frequencies,
which is only legal when IPS tells us we have the headroom. Once that
hack was removed, there was confusion about the driver's frequency
clamping variables: max_delay is the driver's current limit on the
highest frequency the IPS driver wants us to use, while dev_priv->fmax
is the hardware-reported limit that the IPS driver can increase up to.
Tested with IPS driver loaded or not. Note that on Ironlake systems
without the IPS driver loaded this will result in a performance
reduction, and the inital warmup of frequency limits can impact
benchmarking on systems with IPS loaded.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Eric Anholt <eric@anholt.net>
[ickle: demoted a debugging printk]
Cc: stable@kernel.org
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
So long as we adhere to the fence registers rules for alignment and no
overlaps (including with unfenced accesses to linear memory) and account
for the tiled access in our size allocation, we do not have to allocate
the full fenced region for the object. This allows us to fight the bloat
tiling imposed on pre-i965 chipsets and frees up RAM for real use. [Inside
the GTT we still suffer the additional alignment constraints, so it doesn't
magic allow us to render larger scenes without stalls -- we need the
expanded GTT and fence pipelining to overcome those...]
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Like before add a parameter mappable (also to gem_object_pin) and
set it depending upon the context. Only bos that are brought into
the gtt due to an execbuffer call can be put into the unmappable
part of the gtt, everything else (especially pinned objects) need
to be put into the mappable part of the gtt.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Preparing the ringbuffer for adding new commands can fail (a timeout
whilst waiting for the GPU to catch up and free some space). So check
for any potential error before overwriting HEAD with new commands, and
propagate that error back to the user where possible.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Block execbuffer for the fb to be flipped away, not the one that is to
be flipped in.
[ickle: rewritten for -next]
Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Acked-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Acked-by: Kristian Høgsberg <krh@bitplanet.net>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
FDI_PLL_BIOS_0 register is for Ironlake only, don't apply to
Sandybridge.
Original-patch-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The enter argument as implemented by commit 413d45d362 (drm, kdb, kms:
Add an enter argument to mode_set_base_atomic() API) should be more
descriptive as to what it does vs just passing 1 and 0 around.
There is no runtime behavior change as a result of this patch.
Reported-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
CC: David Airlie <airlied@linux.ie>
CC: dri-devel@lists.freedesktop.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
Needed on Ibex Peak and Cougar Point or the panel won't always come on.
Cc: stable@kernel.org
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We do this later (and more properly) when we enable FDI, so we don't
need to do it here.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Wait for vblank after enabling a pipe, make the error messages more
informative, and wait for the pipe to turn off when we disable it.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
CPU eDP needs a different reference clock than PCH eDP, which uses the
standard PCH refclk of 120MHz.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Enable SSC on PCH eDP if possible.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
[ickle: added a posting read of PCH_DREF_CONTROL before the udelay]
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The _DSM method on the integrated graphics device can tell us which
connectors are muxable, so add support for making the call and parsing
out the connector info.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
[ickle: fix compiler warnings for using uninitialized 'result' and
downgrade error message for non-switchable devices]
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
FDI training needs to done and idle for PCH eDP and before we turn the
pipes on, and various eDP checks need to account for PCH attached eDP.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Since we set the output type of PCH attached eDP panels to
INTEL_OUTPUT_eDP this function would never return true when it should.
It's been replaced by working functions.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Currently, if a batch buffer refers to an object with a pending flip,
then we sleep until that pending flip is completed (unpinned and
signalled). This is so that a flip can be queued and the user can
continue rendering to the backbuffer oblivious to whether the buffer is
still pinned as the scan out. (The kernel arbitrating at the last moment
to stall the batch and wait until the buffer is unpinned and replaced as
the front buffer.)
As we only have a queue depth of 1, we can simply wait for the current
pending flip to complete and continue rendering. We can achieve this
with a single WAIT_FOR_EVENT command inserted into the ring buffer prior
to executing the batch, *without* stalling the client.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Some devices such as the radeon chips receive information from user
space which needs to be saved when executing an atomic mode set
operation, else the user space would have to be queried again for the
information.
This patch extends the mode_set_base_atomic() call to pass an argument
to indicate if this is an entry or an exit from an atomic kernel mode
set change. Individual drm drivers can properly save and restore
state accordingly.
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
CC: Jesse Barnes <jbarnes@virtuousgeek.org>
CC: David Airlie <airlied@linux.ie>
CC: dri-devel@lists.freedesktop.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
lockdep spots that the fb_info->lock takes the dev->struct_mutex during
init (due to the device probing) and so we can not hold
dev->struct_mutex when unregistering the framebuffer. Simply reverse the
order of initialisation during cleanup and so do the intel_fbdev_fini()
before the intel_modeset_cleanup.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cancel the output polling work proc before acquiring the struct mutex
to avoid acquiring the work proc mutex with the struct mutex
held. This avoids inverting the lock order seen when the work proc
runs.
Signed-off-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Instead of waiting for the display line value to settle, we can simply
wait for the pipe configuration register 'state' bit to turn off.
Contrarywise, disabling the plane will not cause the display line
value to stop changing, so instead we wait for the vblank interrupt
bit to get set. And, we only do this when we're not about to wait for
the pipe to turn off.
Signed-off-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Avoid cause latencies in other clients by not taking the global struct
mutex and moving the per-client request manipulation a local per-client
mutex. For example, this allows a compositor to schedule a page-flip
(through X) whilst an OpenGL application is monopolising the GPU.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Chris Wilson