On all platforms we now always read the contents of buffers via the GTT,
i.e. using WC cpu access. Reads are slow, but they can be accelerated
with an internal read buffer using sse4.1 (movntqda). This is our
i915_memcpy_from_wc() routine which also checks for sse4.1 support and
so we can fallback to using a regular slow memcpy if we need to.
When compressing the pages, the reads are currently done inside zlib's
fill_window() routine and so we must copy the page into a temporary
which is then already inside the CPU cache and fast for zlib's
compression. When not compressing the pages, we don't need a temporary
and can just use the accelerated read from WC into the destination.
v2: Use zstream locals to reduce diff and allocate the additional
temporary storage only if sse4.1 is supported.
v3: Use length=0 for the sse4.1 support check
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161206124051.17040-1-chris@chris-wilson.co.uk
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Backmerge v4.9-rc8 to get at
commit e94bd1736f
Author: Michel Dänzer <michel.daenzer@amd.com>
Date: Wed Nov 30 17:30:01 2016 +0900
drm: Don't call drm_for_each_crtc with a non-KMS driver
so I can apply Michel's follow-up patch.
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
As we use debugobjects to track the lifetime of fences within our atomic
state, we ideally want to mark those objects as freed along with their
containers. This merits us hookin into config->funcs->atomic_state_free
for this purpose.
This allows us to enable debugobjects for sw-fences without triggering
known issues.
Fixes: fc1584059d ("drm/i915: Integrate i915_sw_fence with debugobjects")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161205142941.21965-4-chris@chris-wilson.co.uk
Soft-pinning depends upon being able to check for availabilty of an
interval and evict overlapping object from a drm_mm range manager very
quickly. Currently it uses a linear list, and so performance is dire and
not suitable as a general replacement. Worse, the current code will oops
if it tries to evict an active buffer.
It also helps if the routine reports the correct error codes as expected
by its callers and emits a tracepoint upon use.
For posterity since the wrong patch was pushed (i.e. that missed these
key points and had known bugs), this is the changelog that should have
been on commit 506a8e87d8 ("drm/i915: Add soft-pinning API for
execbuffer"):
Userspace can pass in an offset that it presumes the object is located
at. The kernel will then do its utmost to fit the object into that
location. The assumption is that userspace is handling its own object
locations (for example along with full-ppgtt) and that the kernel will
rarely have to make space for the user's requests.
This extends the DRM_IOCTL_I915_GEM_EXECBUFFER2 to do the following:
* if the user supplies a virtual address via the execobject->offset
*and* sets the EXEC_OBJECT_PINNED flag in execobject->flags, then
that object is placed at that offset in the address space selected
by the context specifier in execbuffer.
* the location must be aligned to the GTT page size, 4096 bytes
* as the object is placed exactly as specified, it may be used by this
execbuffer call without relocations pointing to it
It may fail to do so if:
* EINVAL is returned if the object does not have a 4096 byte aligned
address
* the object conflicts with another pinned object (either pinned by
hardware in that address space, e.g. scanouts in the aliasing ppgtt)
or within the same batch.
EBUSY is returned if the location is pinned by hardware
EINVAL is returned if the location is already in use by the batch
* EINVAL is returned if the object conflicts with its own alignment (as meets
the hardware requirements) or if the placement of the object does not fit
within the address space
All other execbuffer errors apply.
Presence of this execbuf extension may be queried by passing
I915_PARAM_HAS_EXEC_SOFTPIN to DRM_IOCTL_I915_GETPARAM and checking for
a reported value of 1 (or greater).
v2: Combine the hole/adjusted-hole ENOSPC checks
v3: More color, more splitting, more blurb.
Fixes: 506a8e87d8 ("drm/i915: Add soft-pinning API for execbuffer")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161205142941.21965-2-chris@chris-wilson.co.uk
On my Cherrytrail CUBE iwork8 Air tablet PIPE-A would get stuck on loading
i915 at boot 1 out of every 3 boots, resulting in a non functional LCD.
Once the i915 driver has successfully loaded, the panel can be disabled /
enabled without hitting this issue.
The getting stuck is caused by vlv_init_display_clock_gating() clearing
the DPOUNIT_CLOCK_GATE_DISABLE bit in DSPCLK_GATE_D when called from
chv_pipe_power_well_ops.enable() on driver load, while a pipe is enabled
driving the DSI LCD by the BIOS.
Clearing this bit while DSI is in use is a known issue and
intel_dsi_pre_enable() / intel_dsi_post_disable() already set / clear it
as appropriate.
This commit modifies vlv_init_display_clock_gating() to leave the
DPOUNIT_CLOCK_GATE_DISABLE bit alone fixing the pipe getting stuck.
Changes in v2:
-Replace PIPE-A with "a pipe" or "the pipe" in the commit msg and
comment
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=97330
Cc: stable@vger.kernel.org
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161202142904.25613-1-hdegoede@redhat.com
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
On VLV/CHV some of the watermark values are split across two registers:
low order bits in one, and high order bits in another. So we may not be
able to update a single watermark value atomically, and thus we must be
careful that we don't temporarily introduce out of bounds values during
the reprogramming. To prevent this we can simply zero out all the high
order bits initially, then we update the low order bits, and finally
we update the high order bits with the final value.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1480354637-14209-13-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Before we attempt to turn any planes on or off we must first exit
csxr. That's due to cxsr effectively making the plane enable bits
read-only. Currently we achieve that with a vblank wait right after
toggling the cxsr enable bit. We do the vblank wait even if cxsr was
already off, which seems wasteful, so let's try to only do it when
absolutely necessary.
We could start tracking the cxsr state fully somewhere, but for now
it seems easiest to just have intel_set_memory_cxsr() return the
previous cxsr state.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1480354637-14209-11-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Not validating the mode rate against max. link rate results in not pruning
invalid modes. For e.g, a HBR2 5.4 Gbps 2-lane configuration does not
support 4k@60Hz. But, we do not reject this mode.
So, make use of the helpers in intel_dp to validate mode data rate against
max. link data rate of a configuration.
v3: Renamed local variables again for consistency (Manasi)
v2: Renamed mode data rate local variable to be more explanatory.
Signed-off-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1479243546-17189-1-git-send-email-dhinakaran.pandiyan@intel.com
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
We store DP link rates as link clock frequencies in kHz, just like all
other clock values. But, DP link rates in the DP Spec. are expressed in
Gbps/lane, which seems to have led to some confusion.
E.g., for HBR2
Max. data rate = 5.4 Gbps/lane x 4 lane x 8/10 x 1/8 = 2160000 kBps
where, 8/10 is for channel encoding and 1/8 is for bit to Byte conversion
Using link clock frequency, like we do
Max. data rate = 540000 kHz * 4 lanes = 2160000 kSymbols/s
Because, each symbol has 8 bit of data, this is 2160000 kBps
and there is no need to account for channel encoding here.
But, currently we do 540000 kHz * 4 lanes * (8/10) = 1728000 kBps
Similarly, while computing the required link bandwidth for a mode,
there is a mysterious 1/10 term.
This should simply be pixel_clock kHz * (bpp/8) to give the final result in
kBps
v2: Changed to DIV_ROUND_UP() and comment changes (Ville)
Signed-off-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1479160220-17794-1-git-send-email-dhinakaran.pandiyan@intel.com
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Resync, and we need all the fancy new drm_mm stuff to implement more
efficient evict algorithms for softpin.
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Since the submit/execute split in commit d55ac5bf97 ("drm/i915: Defer
transfer onto execution timeline to actual hw submission") the
global seqno advance was deferred until the submit_request callback.
After wedging the GPU, we were installing a nop_submit_request handler
(to avoid waking up the dead hw) but I had missed converting this over
to the new scheme. Under the new scheme, we have to explicitly call
i915_gem_submit_request() from the submit_request handler to mark the
request as on the hardware. If we don't the request is always pending,
and any waiter will continue to wait indefinitely and hangcheck will not
be able to resolve the lockup.
References: https://bugs.freedesktop.org/show_bug.cgi?id=98748
Testcase: igt/gem_eio/in-flight
Fixes: d55ac5bf97 ("drm/i915: Defer transfer onto execution timeline to actual hw submission")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161122144121.7379-3-chris@chris-wilson.co.uk
(cherry picked from commit 3dcf93f7f2)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Geminilake is mostly backwards compatible with broxton, so change most
of the IS_BROXTON() checks to IS_GEN9_LP(). Differences between the
platforms will be implemented in follow-up patches.
v2: Don't reuse broxton's path in intel_update_max_cdclk().
Don't set plane count as in broxton.
v3: Rebase
v4: Include the check intel_bios_is_port_hpd_inverted().
Commit message.
v5: Leave i915_dmc_info() out; glk's csr version != bxt's. (Rodrigo)
v6: Rebase.
v7: Convert a few mode IS_BROXTON() occurances in pps, ddi, dsi and pll
code. (Rodrigo)
v8: Squash a couple of DDI patches with more conversions. (Rodrigo)
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1480667037-11215-2-git-send-email-ander.conselvan.de.oliveira@intel.com
