DC3CO is useful power state, when DMC detects PSR2 idle frame
while an active video playback, playing 30fps video on 60hz panel
is the classic example of this use case.
B.Specs:49196 has a restriction to enable DC3CO only for Video Playback.
It will be worthy to enable DC3CO after completion of each pageflip
and switch back to DC5 when display is idle because driver doesn't
differentiate between video playback and a normal pageflip.
We will use Frontbuffer flush call tgl_dc3co_flush() to enable DC3CO
state only for ORIGIN_FLIP flush call, because DC3CO state has primarily
targeted for VPB use case. We are not interested here for frontbuffer
invalidates calls because that triggers PSR2 exit, which will
explicitly disable DC3CO.
DC5 and DC6 saves more power, but can't be entered during video
playback because there are not enough idle frames in a row to meet
most PSR2 panel deep sleep entry requirement typically 4 frames.
As PSR2 existing implementation is using minimum 6 idle frames for
deep sleep, it is safer to enable DC5/6 after 6 idle frames
(By scheduling a delayed work of 6 idle frames, once DC3CO has been
enabled after a pageflip).
After manually waiting for 6 idle frames DC5/6 will be enabled and
PSR2 deep sleep idle frames will be restored to 6 idle frames, at this
point DMC will triggers DC5/6 once PSR2 enters to deep sleep after
6 idle frames.
In future when we will enable S/W PSR2 tracking, we can change the
PSR2 required deep sleep idle frames to 1 so DMC can trigger the
DC5/6 immediately after S/W manual waiting of 6 idle frames get
complete.
v2: calculated s/w state to switch over dc3co when there is an
update. [Imre]
Used cancel_delayed_work_sync() in order to avoid any race
with already scheduled delayed work. [Imre]
v3: Cancel_delayed_work_sync() may blocked the commit work.
hence dropping it, dc5_idle_thread() checks the valid wakeref before
putting the reference count, which avoids any chances of dropping
a zero wakeref. [Imre (IRC)]
v4: Used frontbuffer flush mechanism. [Imre]
v5: Used psr.pipe to extract frontbuffer busy bits. [Imre]
Used cancel_delayed_work_sync() in encoder disable path. [Imre]
Used mod_delayed_work() instead of cancelling and scheduling a
delayed work. [Imre]
Used psr.lock in tgl_dc5_idle_thread() to enable psr2 deep
sleep. [Imre]
Removed DC5_REQ_IDLE_FRAMES macro. [Imre]
v6: Used dc3co_exitline check instead of TGL and dc3co allowed_dc_mask
checks, used delayed_work_pending with the psr lock and removed the
psr2_deep_slp_disabled flag. [Imre]
v7: Code refactoring, moved most of functional code to inte_psr.c [Imre]
Using frontbuffer_bits on psr.pipe check instead of
busy_frontbuffer_bits. [Imre]
Calculating dc3co_exit_delay in intel_psr_enable_locked. [Imre]
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Animesh Manna <animesh.manna@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Anshuman Gupta <anshuman.gupta@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191003081738.22101-6-anshuman.gupta@intel.com
DC3CO enabling B.Specs sequence requires to enable end configure
exit scanlines to TRANS_EXITLINE register, programming this register
has to be part of modeset sequence as this can't be change when
transcoder or port is enabled.
When system boots with only eDP panel there may not be real
modeset as BIOS has already programmed the necessary registers,
therefore it needs to force a modeset to enable and configure
DC3CO exitline.
v1: Computing dc3co_exitline crtc state from a DP encoder
compute config. [Imre]
Enabling and disabling DC3CO PSR2 transcoder exitline from
encoder pre_enable and post_disable hooks. [Imre]
Computing dc3co_exitline instead of has_dc3co_exitline bool. [Imre]
v2: Code refactoring for symmetry and to avoid exported function. [Imre]
Removing IS_TIGERLAKE check from compute_config, adding PIPE_A
restriction and clearing dc3co_exitline state if crtc is not active
or it is not PSR2 capable in dc3co exitline compute_config. [Imre]
Using GEN >= 12 check in dc3co exitline get_config. [Imre]
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Animesh Manna <animesh.manna@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Anshuman Gupta <anshuman.gupta@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191003081738.22101-5-anshuman.gupta@intel.com
Add target_dc_state and used by set_target_dc_state API
in order to enable DC3CO state with existing DC states.
target_dc_state will enable/disable the desired DC state in
DC_STATE_EN reg when "DC Off" power well gets disable/enable.
v2: commit log improvement.
v3: Used intel_wait_for_register to wait for DC3CO exit. [Imre]
Used gen9_set_dc_state() to allow/disallow DC3CO. [Imre]
Moved transcoder psr2 exit line enablement from tgl_allow_dc3co()
to a appropriate place haswell_crtc_enable(). [Imre]
Changed the DC3CO power well enabled call back logic as
recommended in review comments. [Imre]
v4: Used wait_for_us() instead of intel_wait_for_reg(). [Imre (IRC)]
v5: using udelay() instead of waiting for DC3CO exit status.
v6: Fixed minor unwanted change.
v7: Removed DC3CO powerwell and POWER_DOMAIN_VIDEO.
v8: Uniform checks by using only target_dc_state instead of allowed_dc_mask
in "DC off" power well callback. [Imre]
Adding "DC off" power well id to older platforms. [Imre]
Removed psr2_deep_sleep flag from tgl_set_target_dc_state. [Imre]
v9: Used switch case for target DC state in
gen9_dc_off_power_well_disable(), checking DC3CO state against
allowed DC mask, using WARN_ON() in
tgl_set_target_dc_state(). [Imre]
v10: Code refactoring and using sanitize_target_dc_state(). [Imre]
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Animesh Manna <animesh.manna@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Anshuman Gupta <anshuman.gupta@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191003081738.22101-4-anshuman.gupta@intel.com
Adding following definition to i915_reg.h
1. DC_STATE_EN register DC3CO bit fields and masks.
DC3CO enable bit will be used by driver to make DC3CO
ready for DMC f/w and status bit will be used as DC3CO
entry status.
2. Transcoder EXITLINE register and its bit fields and mask.
Transcoder EXITLINE enable bit represents PSR2 idle frame
reset should be applied at exit line and exitlines mask
represent required number of scanlines at which DC3CO
exit happens.
B.Specs:49196
v1: Use of REG_BIT and using extra space for EXITLINE_ macro
definition. [Animesh]
v2: Grouping EXITLINE reg bits with EXITLINE(trans) define,
no functional change. [Ville]
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Animesh Manna <animesh.manna@intel.com>
Reviewed-by: Animesh Manna <animesh.manna@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Anshuman Gupta <anshuman.gupta@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191007094607.2111-1-anshuman.gupta@intel.com
The current "disable C3+" workaround for the delayed vblank
irqs on i945gm no longer works. I'm not sure what changed, but
now I need to also disable C2. I also got my hands on a i915gm
machine that suffers from the same issue.
After some furious poking of registers I managed to find a
better workaround: The "Do not Turn off Core Render Clock in C
states" bit. With that I no longer have to disable any C-states,
and as a nice bonus the power cost is only ~1/4 of the
"disable C3+" method (which mind you doesn't even work anymore,
and so would have an even higher power cost if we made it work
by also disabling C2).
So let's throw out all the cpuidle/qos crap and just toggle
the magic bit as needed. And we extend the workaround to cover
i915gm as well.
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191003140231.24408-1-ville.syrjala@linux.intel.com
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
With the introduction of ctx->engines[] we allow multiple logical
contexts to be used on the same engine (e.g. with virtual engines).
According to bspec, aach logical context requires a unique tag in order
for context-switching to occur correctly between them. [Simple
experiments show that it is not so easy to trick the HW into performing
a lite-restore with matching logical IDs, though my memory from early
Broadwell experiments do suggest that it should be generating
lite-restores.]
We only need to keep a unique tag for the active lifetime of the
context, and for as long as we need to identify that context. The HW
uses the tag to determine if it should use a lite-restore (why not the
LRCA?) and passes the tag back for various status identifies. The only
status we need to track is for OA, so when using perf, we assign the
specific context a unique tag.
v2: Calculate required number of tags to fill ELSP.
Fixes: 976b55f0e1 ("drm/i915: Allow a context to define its set of engines")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111895
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191004134015.13204-14-chris@chris-wilson.co.uk
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
Since we cannot allocate underneath the vm->mutex (it is used in the
direct-reclaim paths), we need to shift the allocations off into a
mutexless worker with fence recursion prevention. To know when we need
this protection, we mark up the address spaces that do allocate before
insertion. In the future, we may wish to extend the async bind scheme to
more than just allocations.
v2: s/vm->bind_alloc/vm->bind_async_flags/
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/20191004134015.13204-3-chris@chris-wilson.co.uk
