This patch is to resolve compute hang at resume time.
v2: (agd5f) squash in second fix
Signed-off-by: Sonny Jiang <sonny.jiang@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
In order for hibernation to reliably work we need to properly turn
off the SDMA block, sadly after numerous attemps i haven't not found
proper sequence for clean and full shutdown. So simply reset both
SDMA block, this makes hibernation works reliably on sea island GPU
family (CI)
Hibernation and suspend to ram were tested (several times) on :
Bonaire
Hawaii
Mullins
Kaveri
Kabini
Cc: stable@vger.kernel.org
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
In order for hibernation to reliably work we need to cleanup more
thoroughly the compute ring. Hibernation is different from suspend
resume as when we resume from hibernation the hardware is first
fully initialize by regular kernel then freeze callback happens
(which correspond to a suspend inside the radeon kernel driver)
and turn off each of the block. It turns out we were not cleanly
shutting down the compute ring. This patch fix that.
Hibernation and suspend to ram were tested (several times) on :
Bonaire
Hawaii
Mullins
Kaveri
Kabini
Changed since v1:
- Factor the ring stop logic into a function taking ring as arg.
Cc: stable@vger.kernel.org
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Use kzalloc for allocating one thing rather than
kcalloc(1...
The semantic patch that makes this change is as follows:
// <smpl>
@@
@@
- kcalloc(1,
+ kzalloc(
...)
// </smpl>
Signed-off-by: Maninder Singh <maninder1.s@samsung.com>
Reviewed-by: Vaneet Narang <v.narang@samsung.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Some 855gm models (at least ThinkPad X40) regressed because of
commit b0cd324fae
Author: Jani Nikula <jani.nikula@intel.com>
Date: Wed Nov 12 16:25:43 2014 +0200
drm/i915: don't save/restore backlight hist ctl registers
which tried to make our driver more robust by not blindly saving and
restoring registers, but it failed to take into account
commit 0eb96d6ed3
Author: Jesse Barnes <jbarnes@virtuousgeek.org>
Date: Wed Oct 14 12:33:41 2009 -0700
drm/i915: save/restore BLC histogram control reg across suspend/resume
Fix the regression by enabling hist ctl on gen2.
v2: Improved the comment.
v3: Improved the comment, again.
Reported-and-tested-by: Philipp Gesang <phg@phi-gamma.net>
References: http://mid.gmane.org/20150623222648.GD12335@acheron
Fixes: b0cd324fae ("drm/i915: don't save/restore backlight hist ctl registers")
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: stable@vger.kernel.org
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
The hardware supposedly ignores the WM1 watermarks while the PND
deadline mode is enabled, but clear out the register just in case.
This is what the other OS does, and it does make register dumps look
more consistent when we don't have partial WM1 values lingering in
the registers (some WM1 watermarks already get zeroed when the actually
used DSPFW registers get written).
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Clint Taylor <Clinton.A.Taylor@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Allow tweaking the VLV/CHV memory latencies thorugh sysfs, like we do
for ILK+.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Clint Taylor <Clinton.A.Taylor@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Enabling PM5/DDR DVFS with multiple active pipes isn't a validated
configuration. It does seem to work most of the time at least, but
there is clearly an additional risk of underruns, so let's not play
with fire.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Clint Taylor <Clinton.A.Taylor@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
CxSR (or maxfifo on VLV/CHV) blocks somne changes to the plane control
register (enable bit at least, not quite sure about the rest). So in
order to have the plane enable/disable when we want we need to first
kick the hardware out of cxsr.
Unfortunateloy this requires some extra vblank waits. For the CxSR
enable after the plane update we should eventually use an async
vblank worker, but since we don't have that just do sync vblank
waits. For the disable case we have no choice but to do it
synchronously.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Clint Taylor <Clinton.A.Taylor@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
In order to get decnet memory self refresh residency on VLV, flip it
over to the new CHV way of doing things. VLV doesn't do PM5 or DDR DVFS
so it's a bit simpler.
I'm not sure the currently memory latency used for CHV is really
appropriate for VLV. Some further testing will probably be needed to
figure that out.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Clint Taylor <Clinton.A.Taylor@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Consider which planes are active and compute the FIFO split based on the
relative data rates. Since we only consider the pipe src width rather
than the plane width when computing watermarks it seems best to do the
same when computing the FIFO split as well. This means the only thing we
actually have to consider for the FIFO splut is the bpp, and we can
ignore the rest.
I've just stuffed the logic into the watermark code for now. Eventually
it'll need to move into the atomic update for the crtc.
There's also one extra complication I've not yet considered; Some of the
DSPARB registers contain bits related to multiple pipes. The registers
are double buffered but apparently they update on the vblank of any
active pipe. So doing the FIFO reconfiguration properly when multiple
pipes are active is not going to be fun. But let's ignore that mess for
now.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Clint Taylor <Clinton.A.Taylor@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Turns out the VLV/CHV system agent doesn't understand memory
latencies, so trying to rely on the PND deadline mechanism is not
going to fly especially when DDR DVFS is enabled. Currently we try to
avoid the problems by lying to the system agent about the deadlines
and setting the FIFO watermarks to 8 cachelines. This however leads to
bad memory self refresh residency.
So in order to satosfy everyone we'll just give up on the deadline
scheme and program the watermarks old school based on the worst case
memory latency.
I've modelled this a bit on the ILK+ approach where we compute multiple
sets of watermarks for each pipe (PM2,PM5,DDR DVFS) and when merge thet
appropriate one later with the watermarks from other pipes. There isn't
too much to merge actually since each pipe has a totally independent
FIFO (well apart from the mess with the partially shared DSPARB
registers), but still decopuling the pipes from each other seems like a
good idea.
Eventually we'll want to perform the watermark update in two phases
around the plane update to avoid underruns due to the single buffered
watermark registers. But that's still in limbo for ILK+ too, so I've not
gone that far yet for VLV/CHV either.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Clint Taylor <Clinton.A.Taylor@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Read out the current watermark settings from the hardware at driver init
time. This will allow us to compare the newly calculated values against
the currrent ones and potentially avoid needless WM updates.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Clint Taylor <Clinton.A.Taylor@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Try to update the watermarks on the right side of the plane update. This
is just a temporary hack until we get the proper two part update into
place. However in the meantime this might have some chance of at least
working.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Clint Taylor <Clinton.A.Taylor@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We want cxsr exit to happen ASAP, so toss in some POSTING_READ()s to
make sure things are really kicked off.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Clint Taylor <Clinton.A.Taylor@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We can't elide the fb tracking invalidate if the buffer is already in
the right domain since that would lead to missed screen updates. I'm
pretty sure I've written this already before but must have gotten lost
unfortunately :(
v2: Chris observed that all internal set_domain users already
correctly do the fb invalidate on their own, hence we can move this
just into the set_domain ioctl instead.
v3: I screwed up setting the invalidate ORIGIN_* correctly (Chris).
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reported-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Tested-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We cannot let IPS enabled with no plane on the pipe:
BSpec: "IPS cannot be enabled until after at least one plane has
been enabled for at least one vertical blank." and "IPS must be
disabled while there is still at least one plane enabled on the
same pipe as IPS." This restriction apply to HSW and BDW.
However a shortcut path on update primary plane function
to make primary plane invisible by setting DSPCTRL to 0
was leting IPS enabled while there was no
other plane enabled on the pipe causing flickerings that we were
believing that it was caused by that other restriction where
ips cannot be used when pixel rate is greater than 95% of cdclok.
v2: Don't mess with Atomic path as pointed out by Ville.
v3: Rebase after a long time and atomic path changes.
Accept Ville suggestion of not check !fb
v4: Re-factore on dinq
Reference: https://bugs.freedesktop.org/show_bug.cgi?id=85583
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Tested-by: Kenneth Graunke <kenneth@whitecape.org>
[danvet: Make it compile]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Pull drm updates from Dave Airlie:
"This is the main drm pull request for v4.2.
I've one other new driver from freescale on my radar, it's been posted
and reviewed, I'd just like to get someone to give it a last look, so
maybe I'll send it or maybe I'll leave it.
There is no major nouveau changes in here, Ben was working on
something big, and we agreed it was a bit late, there wasn't anything
else he considered urgent to merge.
There might be another msm pull for some bits that are waiting on
arm-soc, I'll see how we time it.
This touches some "of" stuff, acks are in place except for the fixes
to the build in various configs,t hat I just applied.
Summary:
New drivers:
- virtio-gpu:
KMS only pieces of driver for virtio-gpu in qemu.
This is just the first part of this driver, enough to run
unaccelerated userspace on. As qemu merges more we'll start
adding the 3D features for the virgl 3d work.
- amdgpu:
a new driver from AMD to driver their newer GPUs. (VI+)
It contains a new cleaner userspace API, and is a clean
break from radeon moving forward, that AMD are going to
concentrate on. It also contains a set of register headers
auto generated from AMD internal database.
core:
- atomic modesetting API completed, enabled by default now.
- Add support for mode_id blob to atomic ioctl to complete interface.
- bunch of Displayport MST fixes
- lots of misc fixes.
panel:
- new simple panels
- fix some long-standing build issues with bridge drivers
radeon:
- VCE1 support
- add a GPU reset counter for userspace
- lots of fixes.
amdkfd:
- H/W debugger support module
- static user-mode queues
- support killing all the waves when a process terminates
- use standard DECLARE_BITMAP
i915:
- Add Broxton support
- S3, rotation support for Skylake
- RPS booting tuning
- CPT modeset sequence fixes
- ns2501 dither support
- enable cmd parser on haswell
- cdclk handling fixes
- gen8 dynamic pte allocation
- lots of atomic conversion work
exynos:
- Add atomic modesetting support
- Add iommu support
- Consolidate drm driver initialization
- and MIC, DECON and MIPI-DSI support for exynos5433
omapdrm:
- atomic modesetting support (fixes lots of things in rewrite)
tegra:
- DP aux transaction fixes
- iommu support fix
msm:
- adreno a306 support
- various dsi bits
- various 64-bit fixes
- NV12MT support
rcar-du:
- atomic and misc fixes
sti:
- fix HDMI timing complaince
tilcdc:
- use drm component API to access tda998x driver
- fix module unloading
qxl:
- stability fixes"
* 'drm-next' of git://people.freedesktop.org/~airlied/linux: (872 commits)
drm/nouveau: Pause between setting gpu to D3hot and cutting the power
drm/dp/mst: close deadlock in connector destruction.
drm: Always enable atomic API
drm/vgem: Set unique to "vgem"
of: fix a build error to of_graph_get_endpoint_by_regs function
drm/dp/mst: take lock around looking up the branch device on hpd irq
drm/dp/mst: make sure mst_primary mstb is valid in work function
of: add EXPORT_SYMBOL for of_graph_get_endpoint_by_regs
ARM: dts: rename the clock of MIPI DSI 'pll_clk' to 'sclk_mipi'
drm/atomic: Don't set crtc_state->enable manually
drm/exynos: dsi: do not set TE GPIO direction by input
drm/exynos: dsi: add support for MIC driver as a bridge
drm/exynos: dsi: add support for Exynos5433
drm/exynos: dsi: make use of array for clock access
drm/exynos: dsi: make use of driver data for static values
drm/exynos: dsi: add macros for register access
drm/exynos: dsi: rename pll_clk to sclk_clk
drm/exynos: mic: add MIC driver
of: add helper for getting endpoint node of specific identifiers
drm/exynos: add Exynos5433 decon driver
...
We can't improve a 0 deviation, so when we find such a divider, skip the
remaining ones they won't be better.
This short-circuit the search for 34 of the 373 test frequencies in the
corresponding i-g-t test (tools/skl_compute_wrpll)
v2: Place the short-circuiting code in skl_compute_wrpll() (Paulo)
(I'm sure nobody will notice the spurious removal of a blank line)
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Suggested-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Broxton is using a different register and different bit ordering
for rps status capabilities.
Also GT perf freqency register is different for Broxton so update
that.
Signed-off-by: Bob Paauwe <bob.j.paauwe@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Currently, if an odd divider improves the deviation (minimizes it), we
take that divider. The recommendation is to prefer even dividers.
v2: Move the check at the right place after having inverted the two for
loops in the previous patch.
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The HW validation team came back from further testing with a slightly
changed constraint on the deviation between the DCO frequency and the
central frequency. Instead of +-4%, it's now +1%/-6%.
Unfortunately, the previous algorithm didn't quite cope with these new
constraints, the reason being that it wasn't thorough enough looking at
the possible divider candidates.
The new algorithm looks at all dividers, which is definitely a hammer
approach (we could reduce further the set of dividers to good ones as a
follow up, at the cost of a bit more complicated code). But, at least,
we can now satisfy the +1%/+6% rule for all the "Well known" HDMI
frequencies of my test set (373 entries).
On that subject, the new code is quite extensively tested in
intel-gpu-tools (tools/skl_compute_wrpll).
v2: Fix cycling between central frequencies and dividers (Paulo)
Properly choose the minimal deviation between postive and negative
candidates (Paulo).
On the 373 test frequencies, v2 computes better dividers than v1 (ie
more even dividers and lower deviation on average):
v1: average deviation: 206.52
v2: average deviation: 194.47
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
After Mika's ppgtt cleanup series, all the other free functions have
drm_device as the first parameter, except this one.
No functional changes.
Signed-off-by: Michel Thierry <michel.thierry@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
A safer way to update the PDPx registers is sending lri commands, added
in the ring before the batchbuffer start. Otherwise, the ctx must be idle
before trying to change anything (but the ring-tail) in the ctx image. An
example where the ctx won't be idle is lite-restore.
This patch depends on 5b7e4c9ce ("drm/i915/gtt: Mark TLBS dirty for gen8+").
v2: Combine lri writes (and save 8 commands). (Mika)
v3: Rebase after ring/req changes, and removed references to deprecated patches.
Cc: Dave Gordon <david.s.gordon@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Michel Thierry <michel.thierry@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
If for some reason [1], the page directory/table does not exist, clear_range
would end up in an infinite while loop.
Introduced by commit 06fda602db ("drm/i915: Create page table allocators").
[1] This is already being addressed in one of Mika's patches:
http://mid.gmane.org/1432314314-23530-17-git-send-email-mika.kuoppala@intel.com
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: stable@vger.kernel.org
Reported-by: John Harrison <john.c.harrison@intel.com>
Signed-off-by: Michel Thierry <michel.thierry@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
There is no need for atomicity here. Convert all bitmap
operations to nonatomic variants.
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Michel Thierry <michel.thierry@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Scratch page is part of struct i915_address_space. Move other
scratch entities into the same struct. This is a preparatory patch
for having only one instance of each scratch_pt/pd.
v2: make commit msg more readable
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v1)
[danvet: Bikeshed summary to avoid confusion with vmas.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Write page directory entry without using superfluous
indirect function. Also remove unused device parameter
from the encode function.
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Michel Thierry <michel.thierry@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Dynamic page table allocation might wake the shrinker
when memory is requested for page table structures.
As this happens when we try to allocate the virtual address
during binding, our vma might be among the targets for eviction.
We should do i915_vma_pin() and do pin early in there like Chris
suggests but this is interim solution.
Shield our vma from shrinker by incrementing pin count before
the virtual address is allocated.
The proper place to fix this would be in gem, inside of
i915_vma_pin(). But we don't have that yet so take the short
cut as a intermediate solution.
Testcase: igt/gem_ctx_thrash
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Michel Thierry <michel.thierry@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Lay out scratch page structure in similar manner than other
paging structures. This allows us to use the same tools for
setup and teardown.
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Michel Thierry <michel.thierry@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Make paging structure type agnostic *_px macros to access
page dma struct, the backing page and the dma address.
This makes the code less cluttered on internals of
i915_page_dma.
v2: Superfluous const -> nonconst removed
v3: Rebased
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Michel Thierry <michel.thierry@intel.com> (v2)
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
As there is flushing involved when we have done the cpu
write, make functions for mapping for cpu space. Make macros
to map any type of paging structure.
v2: Make it clear tha flushing kunmap is only for ppgtt (Ville)
v3: Flushing fixed (Ville, Michel). Removed superfluous semicolon
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Michel Thierry <michel.thierry@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
When we setup page directories and tables, we point the entries
to a to the next level scratch structure. Make this generic
by introducing a fill_page_dma which maps and flushes. We also
need 32 bit variant for legacy gens.
v2: Fix flushes and handle valleyview (Ville)
v3: Now really fix flushes (Michel, Ville)
Reviewed-by: Michel Thierry <michel.thierry@intel.com>
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
This has slipped in somewhere but it was harmless
as we check the page pointer before teardown.
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Michel Thierry <michel.thierry@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
All the paging structures are now similar and mapped for
dma. The unmapping is taken care of by common accessors, so
don't overload the reader with such details.
v2: Be consistent with goto labels (Michel)
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Michel Thierry <michel.thierry@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
All our paging structures have struct page and dma address
for that page.
Add struct for page/dma address pairs and use it to make
the setup and teardown for different paging structures
identical.
Include the page directory offset also in the struct for legacy
gens. Rename it to clearly point out that it is offset into the
ggtt.
v2: Add comment about ggtt_offset (Michel)
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Michel Thierry <michel.thierry@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
