- 19 May, 2020 15 commits
-
-
Chris Wilson authored
We recorded the execlists->queue_priority_hint update for the inflight request without kicking the tasklet. The next submitted request then failed to be scheduled as it had a lower priority than the hint, leaving the HW running with only the inflight request. Fixes: 6cebcf74 ("drm/i915: Tweak scheduler's kick_submission()") Signed-off-by:
Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by:
Mika Kuoppala <mika.kuoppala@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20200519063123.20673-1-chris@chris-wilson.co.uk
-
Chris Wilson authored
A useful metric of the system's health is how fast we can tell the GPU to do various actions, so measure our latency. v2: Refactor all the instruction building into emitters. v3: Mark the error handling if not perfect, at least consistent. Signed-off-by:
-
Pankaj Bharadiya authored
struct drm_device specific drm_WARN* macros include device information in the backtrace, so we know what device the warnings originate from. Prefer drm_WARN* over WARN*. Conversion is done with below semantic patch: @@ identifier func, T; @@ func(struct intel_runtime_pm *T,...) { + struct drm_i915_private *i915 = container_of(T, struct drm_i915_private, runtime_pm); <+... ( -WARN( +drm_WARN(&i915->drm, ...) | -WARN_ON( +drm_WARN_ON(&i915->drm, ...) | -WARN_ONCE( +drm_WARN_ONCE(&i915->drm, ...) | -WARN_ON_ONCE( +drm_WARN_ON_ONCE(&i915->drm, ...) ) ...+> } Signed-off-by:Pankaj Bharadiya <pankaj.laxminarayan.bharadiya@intel.com> Signed-off-by:
Jani Nikula <jani.nikula@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20200504181600.18503-10-pankaj.laxminarayan.bharadiya@intel.com
-
Pankaj Bharadiya authored
struct drm_device specific drm_WARN* macros include device information in the backtrace, so we know what device the warnings originate from. Prefer drm_WARN_ON over WARN_ON. Conversion is done with below sementic patch: @@ identifier func, T; @@ func(...) { ... struct intel_crtc *T = ...; +struct drm_i915_private *dev_priv = to_i915(T->base.dev); <+... -WARN_ON( +drm_WARN_ON(&dev_priv->drm, ...) ...+> } @@ identifier func, T; @@ func(struct intel_crtc_state *T,...) { +struct drm_i915_private *dev_priv = to_i915(T->uapi.crtc->dev); <+... -WARN_ON( +drm_WARN_ON(&dev_priv->drm, ...) ...+> } changes since v1: - Added dev_priv local variable and used it in drm_WARN_ON calls (Jani) Signed-off-by: -
Pankaj Bharadiya authored
struct drm_device specific drm_WARN* macros include device information in the backtrace, so we know what device the warnings originate from. Prefer drm_WARN_ON over WARN_ON. Signed-off-by:
-
Pankaj Bharadiya authored
struct drm_device specific drm_WARN* macros include device information in the backtrace, so we know what device the warnings originate from. Prefer drm_WARN_ON over WARN_ON. changes since v1: - Add parentheses around the dev_priv macro argument (Jani) Signed-off-by:
-
Pankaj Bharadiya authored
struct drm_device specific drm_WARN* macros include device information in the backtrace, so we know what device the warnings originate from. Prefer drm_WARN* over WARN* at places where struct drm_device pointer can be extracted. Signed-off-by:
-
Pankaj Bharadiya authored
struct drm_device specific drm_WARN* macros include device information in the backtrace, so we know what device the warnings originate from. Prefer drm_WARN_ON over WARN_ON. Conversion is done with below sementic patch: @@ identifier func, T; @@ func(...) { ... struct drm_i915_private *T = ...; <+... -WARN_ON( +drm_WARN_ON(&T->drm, ...) ...+> } @@ identifier func, T; @@ func(struct intel_digital_port *T,...) { +struct drm_i915_private *i915 = to_i915(T->base.base.dev); <+... -WARN_ON( +drm_WARN_ON(&i915->drm, ...) ...+> } changes since v1: - Add i915 local variable and use it in drm_WARN_ON (Jani) Signed-off-by: -
Pankaj Bharadiya authored
struct drm_device specific drm_WARN* macros include device information in the backtrace, so we know what device the warnings originate from. Prefer drm_WARN* over WARN* calls. changes since v1: - Added dev_priv local variable and used it in drm_WARN* calls (Jani) Signed-off-by:
-
Pankaj Bharadiya authored
struct drm_device specific drm_WARN* macros include device information in the backtrace, so we know what device the warnings originate from. Prefer drm_WARN* over WARN* at places where struct intel_dp or struct drm_i915_private pointer is available. Conversion is done with below sementic patch: @rule1@ identifier func, T; @@ func(...) { ... struct drm_i915_private *T = ...; <+... ( -WARN_ON( +drm_WARN_ON(&T->drm, ...) | -WARN_ON_ONCE( +drm_WARN_ON_ONCE(&T->drm, ...) ) ...+> } @rule2@ identifier func, T; @@ func(struct drm_i915_private *T,...) { <+... ( -WARN_ON( +drm_WARN_ON(&T->drm, ...) | -WARN_ON_ONCE( +drm_WARN_ON_ONCE(&T->drm, ...) ) ...+> } @rule3@ identifier func, T; @@ func(struct intel_dp *T,...) { + struct drm_i915_private *i915 = dp_to_i915(T); <+... ( -WARN_ON( +drm_WARN_ON(&i915->drm, ...) | -WARN_ON_ONCE( +drm_WARN_ON_ONCE(&i915->drm, ...) ) ...+> } @rule4@ identifier func, T; @@ func(...) { ... struct intel_dp *T = ...; + struct drm_i915_private *i915 = dp_to_i915(T); <+... ( -WARN_ON( +drm_WARN_ON(&i915->drm, ...) | -WARN_ON_ONCE( +drm_WARN_ON_ONCE(&i915->drm, ...) ) ...+> } Signed-off-by: -
Pankaj Bharadiya authored
struct drm_device specific drm_WARN* macros include device information in the backtrace, so we know what device the warnings originate from. Prefer drm_WARN_ON over WARN_ON at places where struct i915_power_domains struct is available. Conversion is done with below sementic patch: @@ identifier func, T; @@ func(struct i915_power_domains *T,...) { + struct drm_i915_private *i915 = container_of(T, struct drm_i915_private, power_domains); <+... -WARN_ON( +drm_WARN_ON(&i915->drm, ...) ...+> } changes since v1: - Fix commit subject (Jani) Signed-off-by: -
Imre Deak authored
Make sure to select the port's AUX power domain while holding the TC port lock. The domain depends on the port's current TC mode, which may get changed under us if we're not holding the lock. This was left out from commit 8c10e226 ("drm/i915: Keep the TypeC port mode fixed for detect/AUX transfers") Cc: <stable@vger.kernel.org> # v5.4+ Signed-off-by:
Imre Deak <imre.deak@intel.com> Reviewed-by:
José Roberto de Souza <jose.souza@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20200514204553.27193-1-imre.deak@intel.com
-
Ville Syrjälä authored
I've checked a bunch of gen3/4 machines and all seem to have consistent FSB frequency information in the CLKCFG register. So let's read out hrawclk on all gen3+ machines. Although apart from g4x/pnv aux/pps dividers we only really need this for for i965g/gm cs timestamp increment. The CLKCFG memory clock values seem less consistent but we don't care about those here. For posterity here's a list of CLKCFG vs. FSB dumps from a bunch of machines (only missing lpt for a full set): machine CLKCFG FSB alv1 0x00001411 533 alv2 0x00000420 400 (Chris) gdg1 0x20000022 800 gdg2 0x20000022 800 cst 0x00010043 666 blb 0x00002034 1333 pnv1 0x00000423 666 pnv2 0x00000433 666 965gm 0x00004342 800 946gz 0x00000022 800 965g 0x00000422 800 g35 0x00000430 1066 0x00000434 1333 ctg1 0x00644056 1066 ctg2 0x00644066 1066 elk1 0x00012420 1066 0x00012424 1333 0x00012436 1600 0x00012422 800 elk2 0x00012040 1066 For the mobile parts the chipset docs generally have these documented to some degree (alv being the exception). The two settings w/o any evidence are 0x5=400MHz on desktop and 0x7=1333MHz on mobile. Though the mobile 1333MHz case probably doesn't even exist since ctg is only documented to go up to 1066MHz. v2: Fix 400mhz readout for Chris's alv/celeron machine Do a clean mobile vs. dekstop split since that's really what seems to be going on Cc: Chris Wilson <chris@chris-wilson.co.uk> Cc: Lionel Landwerlin <lionel.g.landwerlin@intel.com> Signed-off-by:Ville Syrjälä <ville.syrjala@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20200514123838.3017-3-ville.syrjala@linux.intel.comAcked-by:
-
Ville Syrjälä authored
Document the fact that we aren't reading out the actual FSB frequency but rather just the state of the FSB straps. Some BIOSen allow you to configure the two independently. So if someone sets the two up in an inconsistent manner we'll get the wrong answer here and thus will end up with incorrect aux/pps clock dividers. Alas, proper docs are no longer around so we can't do any better. Signed-off-by:
-
Ville Syrjälä authored
Looks like elk redefines some of the CLKCFG FSB values to make room for 400 MHz FSB. The setting overlaps with one of the 266MHz settings (which is even documented in the ctg docs, and cofirmed to be correct on my ctg). So we limit the special case to elk only. Though it might also be that we have some kind of desktop vs. mobile difference going on here as eg. both g35 and elk use 0x0 for the 266 MHz setting, vs. 0x6 used by ctg). The g35 doesn't let me select 400MHz for the FSB strap so can't confirm which way it would go here. But anyways as it seems only elk has the 400MHz option we shouldn't lose anything by limiting the special case to it alone. My earlier experiments on this appear to have been nonsense as the comment I added claims that FSB strap of 400MHz results in a value of 0x4, but I've now retested it and I definitely get a value of 0x6 instead. So let's remove that bogus comment. v2: s/_ELK/_ALT/ in the define in anticipation of a full mobile vs. desktop CLKCFG split Signed-off-by:
-
- 18 May, 2020 4 commits
-
-
Chris Wilson authored
drivers/gpu/drm/i915/intel_pm.c:5920 skl_ddb_add_affected_pipes() error: uninitialized symbol 'ret'. Fixes: 3cf43cdc ("drm/i915: Introduce proper dbuf state") Signed-off-by:
-
Ville Syrjälä authored
The current dbuf slice computation only happens when there are active pipes. If we are turning off all the pipes we just leave the dbuf slice mask at it's previous value, which may be something other that BIT(S1). If runtime PM will kick in it will however turn off everything but S1. Then on the next atomic commit (if the new dbuf slice mask matches the stale value we left behind) the code will not turn on the other slices we now need. This will lead to underruns as the planes are trying to use a dbuf slice that's not powered up. To work around let's just just explicitly set the dbuf slice mask to BIT(S1) when we are turning off all the pipes. Really the code should just calculate this stuff the same way regardless whether the pipes are on or off, but we're not quite there yet (need a bit more work on the dbuf state for that). v2: Let's not put the fix into dead code Cc: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com> Acked-by:
-
Chris Wilson authored
In order to keep all the tasklets in the same execution lists and so fifo ordered, be consistent and use the same priority for all. Signed-off-by:
Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20200518081440.17948-3-chris@chris-wilson.co.uk
-
Chris Wilson authored
Tvrtko spotted that some selftests were using 'break' not 'continue', which will fail for discontiguous engine layouts such as on Icelake (which may have vcs0 and vcs2). Reported-by:
-
- 16 May, 2020 2 commits
-
-
Nathan Chancellor authored
When CONFIG_DRM_I915_DEBUG_GEM is not set, clang warns: drivers/gpu/drm/i915/gvt/scheduler.c:884:1: warning: function 'check_shadow_context_ppgtt' is not needed and will not be emitted [-Wunneeded-internal-declaration] check_shadow_context_ppgtt(struct execlist_ring_context *c, struct intel_vgpu_mm *m) ^ 1 warning generated. This warning is similar to -Wunused-function but rather than warning that the function is completely unused, it warns that it is used in some expression within the file but that expression will be evaluated to a constant or be optimized away in the final assembly, essentially making it appeared used but really isn't. Usually, this happens when a function or variable is only used in sizeof, where it will appear to be used but will be evaluated at compile time and not be required to be emitted. In this case, the function is only used in GEM_BUG_ON, which is defined as BUILD_BUG_ON_INVALID, which intentionally follows this pattern. To fix this warning, add __maybe_unused to make it clear that this is intentional depending on the configuration. Fixes: bec3df93 ("drm/i915/gvt: Support PPGTT table load command") Link: https://github.com/ClangBuiltLinux/linux/issues/1027Signed-off-by:
Nathan Chancellor <natechancellor@gmail.com> Reviewed-by:
-
Chris Wilson authored
As we no longer use the shmemfs allocation directly, we do not expect to receive -ENOSPC from a backing store allocation. The potential sources for -ENOSPC are then our own internal eviction code, so the choice is either to kill the potential application with SIGBUS or to retry the faulthandler. In this patch we retry the fault handler, but since this is a should never happen condition, it is arguable that we gather up copious debug and kill the application. At worst, we cause an interruptible busy-wait, stalling the application -- all causes should be transient and the system should eventually recover. A small stall is hopefully a better outcome than random oomkiller. Signed-off-by:
Matthew Auld <matthew.auld@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20200515200031.12034-1-chris@chris-wilson.co.uk
-
- 15 May, 2020 14 commits
-
-
Ville Syrjälä authored
Combine the two per-pipe dbuf debugs into one, and use the canonical [CRTC:%d:%s] style to identify the crtc. Also use the same style as the plane code uses for the ddb start/end, and prefix bitmask properly with 0x to make it clear they are in fact bitmasks. The "how many total slices we are going to use" debug we move to outside the crtc loop so it gets printed only once at the end. Cc: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com> Signed-off-by:
Stanislav Lisovskiy <stanislav.lisovskiy@intel.com>
-
Ville Syrjälä authored
Encapsulate the dbuf state more by moving the pre/post plane functions out from intel_display.c. We stick them into intel_pm.c since that's where the rest of the code lives for now. Eventually we should add a new file for this stuff at which point we also need to decide if it makes sense to even split the wm code from the ddb code, or to keep them together. Cc: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com> Signed-off-by:
-
Ville Syrjälä authored
skl_ddb_get_hw_state() is redundant and kinda called in thw wrong spot anyway. Just kill it. Cc: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com> Signed-off-by:
-
Ville Syrjälä authored
Add a global state to track the dbuf slices. Gets rid of all the nasty coupling between state->modeset and dbuf recomputation. Also we can now totally nuke state->active_pipe_changes. dev_priv->wm.distrust_bios_wm still remains, but that too will get nuked soon. Cc: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com> Signed-off-by:
-
Ville Syrjälä authored
The low level dbuf slice code is rather inconsitent with its functiona naming and organization. Make it more consistent. Also share the enable/disable functions between all platforms since the same code works just fine for all of them. Cc: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com> Signed-off-by:
-
Ville Syrjälä authored
Polish some of the dbuf code to give more meaningful debug messages and whatnot. Also we can switch over to the per-device debugs/warns at the same time. Cc: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com> Signed-off-by:
-
Ville Syrjälä authored
Currently skl_compute_dbuf_slices() returns 0 for any inactive pipe on icl+, but returns BIT(S1) on pre-icl for any pipe (whether it's active or not). Let's make the behaviour consistent and always return 0 for any inactive pipe. Cc: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com> Signed-off-by:
-
Arnd Bergmann authored
After the function is no longer marked 'inline', there is now a new warning pointing out that the only caller is inside of an #ifdef: drivers/gpu/drm/i915/display/intel_panel.c:493:12: warning: 'scale_user_to_hw' defined but not used [-Wunused-function] 493 | static u32 scale_user_to_hw(struct intel_connector *connector, | ^~~~~~~~~~~~~~~~ Move the function itself into that #ifdef as well. Fixes: 81b55ef1 ("drm/i915: drop a bunch of superfluous inlines") Signed-off-by:Arnd Bergmann <arnd@arndb.de> Reviewed-by:
-
Ville Syrjälä authored
Put the customary () around the macro argument in the overlay colorkey macros. And while at switch to using a consistent case for the hex constants. Signed-off-by:
-
Ville Syrjälä authored
We pass the plane data through the pipe gamma for all the other planes. Can't see why we should treat the overlay differently, so let's enable pipe gamma for it as well. Signed-off-by:
-
Ville Syrjälä authored
Put the overlay color conversion unit into 10bit mode if the pipe isn't using the 8bit legacy gamma. Not 100% sure this is what the intention of the bit was but makes at least some sense to me. Signed-off-by:
-
Ville Syrjälä authored
As with the video sprites the colorkey is always specified as 8bpc. For 10bpc primary plane formats we just ignore the two lsbs of each component. For C8 we'll replicate the same key to each chanel, which is what the hardware wants. Signed-off-by:
-
Ville Syrjälä authored
Apparently the 128x128 and 256x256 ARGB cursor modes were only added on LPT/CST. While the display section of bspec isn't super clear on the subject, it does highlight these two modes in a different color, has a few changlog entries indicating the 256x256 mode was added for a LPT DCN, and that the 128x128 mode was also added later (though no DCN/platform note there). The "device dependencies" bspec section does list the 256x256x32 as a new feature for LPT/CST, and goes on to mention that current hw only has the 64x64x32 mode (which reinforces the notion that the 128x128 mode was also added at the same time). Testing on actual hardware confirms all of this. CI shows all the 128x128 and 256x256 tests failing on GDG, and my ALV definitely doesn't like them. So we shall limit GDG/ALV to 64x64 only. And while at it let's adjust the mobile gen2 case to list the two platforms explicitly so that the if-ladder looks reasonably uniform. Signed-off-by:
-
Joonas Lahtinen authored
Signed-off-by:Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
-
- 14 May, 2020 5 commits
-
-
Joonas Lahtinen authored
To ensure cross-driver locking compatibility, document the expected guidelines for implementing the GEM locking in i915. Note that this is a description of how things should end up after being reworked, and does not reflect the current state of things. v2: Use rst note:: tag (Rodrigo) Signed-off-by:
Daniel Vetter <daniel.vetter@intel.com> Signed-off-by:
Rodrigo Vivi <rodrigo.vivi@intel.com> Acked-by:
Dave Airlie <airlied@redhat.com> Link: https://patchwork.freedesktop.org/patch/msgid/20190830105053.17491-1-joonas.lahtinen@linux.intel.com
-
Ville Syrjälä authored
Pull the code to do the CS timestamp ns<->ticks conversion into helpers and use them all over. The check in i915_perf_noa_delay_set() seems a bit dubious, so we switch it to do what I assume it wanted to do all along (ie. make sure the resulting delay in CS timestamp ticks doesn't exceed 32bits)? Cc: Lionel Landwerlin <lionel.g.landwerlin@intel.com> Signed-off-by:
-
Ville Syrjälä authored
kHz isn't accurate enough for storing the CS timestamp frequency on some of the platforms. Store the value in Hz instead. Cc: Lionel Landwerlin <lionel.g.landwerlin@intel.com> Signed-off-by:
Lionel Landwerlin <lionel.g.landwerlin@intel.com>
-
Ville Syrjälä authored
Bunch of places use a 64bit divisor needlessly. Switch to 32bit divisor. Cc: Lionel Landwerlin <lionel.g.landwerlin@intel.com> Signed-off-by:
-
Stanislav Lisovskiy authored
Flip the switch and enable SAGV support for Gen12 also. Signed-off-by:
-
