Forgo the struct_mutex serialisation for i915_active, and interpose its
own mutex handling for active/retire.
This is a multi-layered sleight-of-hand. First, we had to ensure that no
active/retire callbacks accidentally inverted the mutex ordering rules,
nor assumed that they were themselves serialised by struct_mutex. More
challenging though, is the rule over updating elements of the active
rbtree. Instead of the whole i915_active now being serialised by
struct_mutex, allocations/rotations of the tree are serialised by the
i915_active.mutex and individual nodes are serialised by the caller
using the i915_timeline.mutex (we need to use nested spinlocks to
interact with the dma_fence callback lists).
The pain point here is that instead of a single mutex around execbuf, we
now have to take a mutex for active tracker (one for each vma, context,
etc) and a couple of spinlocks for each fence update. The improvement in
fine grained locking allowing for multiple concurrent clients
(eventually!) should be worth it in typical loads.
v2: Add some comments that barely elucidate anything :(
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191004134015.13204-6-chris@chris-wilson.co.uk
Replace the struct_mutex requirement for pinning the i915_vma with the
local vm->mutex instead. Note that the vm->mutex is tainted by the
shrinker (we require unbinding from inside fs-reclaim) and so we cannot
allocate while holding that mutex. Instead we have to preallocate
workers to do allocate and apply the PTE updates after we have we
reserved their slot in the drm_mm (using fences to order the PTE writes
with the GPU work and with later unbind).
In adding the asynchronous vma binding, one subtle requirement is to
avoid coupling the binding fence into the backing object->resv. That is
the asynchronous binding only applies to the vma timeline itself and not
to the pages as that is a more global timeline (the binding of one vma
does not need to be ordered with another vma, nor does the implicit GEM
fencing depend on a vma, only on writes to the backing store). Keeping
the vma binding distinct from the backing store timelines is verified by
a number of async gem_exec_fence and gem_exec_schedule tests. The way we
do this is quite simple, we keep the fence for the vma binding separate
and only wait on it as required, and never add it to the obj->resv
itself.
Another consequence in reducing the locking around the vma is the
destruction of the vma is no longer globally serialised by struct_mutex.
A natural solution would be to add a kref to i915_vma, but that requires
decoupling the reference cycles, possibly by introducing a new
i915_mm_pages object that is own by both obj->mm and vma->pages.
However, we have not taken that route due to the overshadowing lmem/ttm
discussions, and instead play a series of complicated games with
trylocks to (hopefully) ensure that only one destruction path is called!
v2: Add some commentary, and some helpers to reduce patch churn.
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: https://patchwork.freedesktop.org/patch/msgid/20191004134015.13204-4-chris@chris-wilson.co.uk
Link training is failling when running link at 2.7GHz and 1.62GHz and
following BSpec pll algorithm.
Comparing the values calculated and the ones from the reference table
it looks like MG_CLKTOP2_CORECLKCTL1_A_DIVRATIO should not always set
to 5. For DP ports ICL mg pll algorithm sets it to 10 or 5 based on
div2 value, that matches with dkl hardcoded table.
So implementing this way as it proved to work in HW and leaving a
comment so we know why it do not match BSpec.
v4:
Using the same is_dp check as ICL, need testing on HDMI over tc port
Issue reported on BSpec 49204.
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190926210659.56317-3-jose.souza@intel.com
Added DKL Phy sequences and helpers functions to program voltage
swing, clock gating and dp mode.
It is not written in DP enabling sequence but "PHY Clockgating
programming" states that clock gating should be enabled after the
link training but doing so causes all the following trainings to fail
so not enabling it for.
v2:
Setting the right HIP_INDEX_REG bits (José)
v3:
Adding the meaning of each column of tgl_dkl_phy_ddi_translations
Adding if gen >= 12 on intel_ddi_hdmi_level() and
intel_ddi_pre_enable_hdmi() instead of reuse part of gen >= 11 if
v4:
Moved the DP_MODE lane programing to another patch as ICL also
needed it
Sharing icl_phy_set_clock_gating() and icl_program_mg_dp_mode() with
TGL as bits and programing as now it almost identical to ICL
BSpec: 49292
BSpec: 49190
Cc: Imre Deak <imre.deak@intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Signed-off-by: Clinton A Taylor <clinton.a.taylor@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190926210659.56317-2-jose.souza@intel.com
BSpec was updated(r146548) with a new MG_DP_MODE Programming table,
now taking in consideration the pin assignment and allowing us to
optimize power by shutting down available but not needed lanes.
It was tested on ICL and TGL, with adaptors that used pin assignment
C and B, reversing the connector and going to different modes testing
the not needed lane shutdown.
v5:
Using crtc_state->lane_count instead of dp.lane_count
BSpec: 21735
BSpec: 49292
Cc: Imre Deak <imre.deak@intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Clinton A Taylor <clinton.a.taylor@intel.com>
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190926210659.56317-1-jose.souza@intel.com
According to the bspec, GLK/CNL have a smaller small joiner RAM buffer
than ICL+. This feels like something that could easily change again on
future platforms, so let's just add a function to return the proper
per-platform buffer size. That may also slightly simplify the upcoming
bigjoiner enabling.
Since we have to change intel_dp_dsc_get_output_bpp()'s signature to
pass the dev_priv down for the platform check, let's take the
opportunity to also make that function static since it isn't used
outside the intel_dp file.
v2: Minor rebase on top of Maarten's changes.
Bspec: 20388
Bspec: 49259
Cc: Manasi Navare <manasi.d.navare@intel.com>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Manasi Navare <manasi.d.navare@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190925234542.24289-1-matthew.d.roper@intel.com
There was a integer wraparound when mode_clock became too high,
and we didn't correct for the FEC overhead factor when dividing,
with the calculations breaking at HBR3.
As a result our calculated bpp was way too high, and the link width
limitation never came into effect.
Print out the resulting bpp calcululations as a sanity check, just
in case we ever have to debug it later on again.
We also used the wrong factor for FEC. While bspec mentions 2.4%,
all the calculations use 1/0.972261, and the same ratio should be
applied to data M/N as well, so use it there when FEC is enabled.
This fixes the FIFO underrun we are seeing with FEC enabled.
Changes since v2:
- Handle fec_enable in intel_link_compute_m_n, so only data M/N is adjusted. (Ville)
- Fix initial hardware readout for FEC. (Ville)
Changes since v3:
- Remove bogus fec_to_mode_clock. (Ville)
Changes since v4:
- Use the correct register for icl. (Ville)
- Split hw readout to a separate patch.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Fixes: d9218c8f6c ("drm/i915/dp: Add helpers for Compressed BPP and Slice Count for DSC")
Cc: <stable@vger.kernel.org> # v5.0+
Cc: Manasi Navare <manasi.d.navare@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190925082110.17439-1-maarten.lankhorst@linux.intel.com
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
If platform supports and has modular FIA is enabled, the registers
bits also change, example: reading TC3 registers with modular FIA
enabled, driver should read from FIA2 but with TC1 bits offsets.
It is described in BSpec 50231 for DFLEXDPSP, other registers don't
have the BSpec description but testing in real hardware have proven
that it had moved for all other registers too.
v2:
- Caching index in tc_phy_fia_idx, instead of calculate it each time
v3:
- Setting tc_phy_fia and tc_phy_fia_idx in the same function
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190920205810.211048-3-jose.souza@intel.com
Gamma lut programming can be programmed using DSB
where bulk register programming can be done using indexed
register write which takes number of data and the mmio offset
to be written.
Currently enabled for 12-bit gamma LUT which is enabled by
default and later 8-bit/10-bit will be enabled in future
based on need.
v1: Initial version.
v2: Directly call dsb-api at callsites. (Jani)
v3:
- modified the code as per single dsb instance per crtc. (Shashank)
- Added dsb get/put call in platform specific load_lut hook. (Jani)
- removed dsb pointer from dev_priv. (Jani)
v4: simplified code by dropping ref-count implementation. (Shashank)
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Shashank Sharma <shashank.sharma@intel.com>
Signed-off-by: Animesh Manna <animesh.manna@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20190920115930.27829-9-animesh.manna@intel.com
