Update GuC ADS size allocation to include space for
the lists of error state capture register descriptors.
Then, populate GuC ADS with the lists of registers we want
GuC to report back to host on engine reset events. This list
should include global, engine-class and engine-instance
registers for every engine-class type on the current hardware.
Ensure we allocate a persistent store for the register lists
that are populated into ADS so that we don't need to allocate
memory during GT resets when GuC is reloaded and ADS population
happens again.
NOTE: Start with a sample static table of register lists to
layout the framework before adding real registers in subsequent
patch. This static register tables are a different format from
the ADS populated list.
Signed-off-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220321164527.2500062-2-alan.previn.teres.alexis@intel.com
Use drm_clflush_virt_range instead of directly invoking clflush. This
will prevent compiler errors when building for non-x86 architectures.
v2(Michael Cheng): Remove extra clflush
v3(Michael Cheng): Remove memory barrier since drm_clflush_virt_range
takes care of it.
v4(Michael Cheng): Get the size of value and not the size of the pointer
when passing in execlists->csb_write. Thanks to Matt
Roper for pointing this out.
Signed-off-by: Michael Cheng <michael.cheng@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220321223819.72833-4-michael.cheng@intel.com
Drop invalidate_csb_entries and directly call drm_clflush_virt_range.
This allows for one less function call, and prevent complier errors when
building for non-x86 architectures.
v2(Michael Cheng): Drop invalidate_csb_entries function and directly
invoke drm_clflush_virt_range. Thanks to Tvrtko for the
sugguestion.
v3(Michael Cheng): Use correct parameters for drm_clflush_virt_range.
Thanks to Tvrtko for pointing this out.
v4(Michael Cheng): Simplify &execlists->csb_status[0] to
execlists->csb_status. Thanks to Matt Roper for the
suggestion.
Signed-off-by: Michael Cheng <michael.cheng@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220321223819.72833-3-michael.cheng@intel.com
Make the dbuf bandwidth min cdclk calculations match the spec
more closely. Supposedly the arbiter can only guarantee an equal
share of the total bandwidth of the slice to each active plane
on that slice. So we take the max bandwidth of any of the planes
on each slice and multiply that by the number of active planes
on the slice to get a worst case estimate on how much bandwidth
we require.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220303191207.27931-9-ville.syrjala@linux.intel.com
Reviewed-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com>
There's really no need to maintain these total[] arrays to
track the size of each plane's ddb allocation. We just stick
the results straight into the crtc_state ddb tracking structures.
The main annoyance with all this is the mismatch between
wm_uv vs. ddb_y on pre-icl. If only the hw was consistent in
what it considers the primary source of information we could
avoid some of the uglyness. But since that is not the case
we need a bit of special casing for planar formats.
v2: Keep the ddb entry zeroed when the plane is disabled
Reviewed-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220303191207.27931-5-ville.syrjala@linux.intel.com
Handle the plane relative data rate in exactly the same
way as we already handle the real data rate. Ie. pre-calculate
it during intel_plane_atomic_check_with_state(), and assign/clear
it for the Y plane as needed. This should guarantee that the
tracking is 100% consistent, and makes me have to think less
when the same apporach is used by both types of data rate.
We might even want to consider replacing the relative
data rate with the real data rate entirely, but it's not
clear if that will produce less optimal plane ddb
allocations. So for now lets keep using the current approach.
v2: Rebase due to async flip wm optimization
Reviewed-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220303191207.27931-4-ville.syrjala@linux.intel.com
Let's store the plane allocation in a manner which more closely
matches how the hw operates. That is, we store the packed/CbCr
ddb in one struct, and the Y ddb in another. Currently we're
storing packed/Y in one struct, CbCr in the other.
This also works pretty well for icl+ where the UV plane is
the main plane and the Y plane is subservient to it. Although
in this case we do not even use ddb_y as we do the ddb allocation
in terms of hw planes.
v2: Rebase
Reviewed-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220303191207.27931-2-ville.syrjala@linux.intel.com
On platforms capable of allowing 8K (7680 x 4320) modes, pinning 2 or
more framebuffers/scanout buffers results in only one that is mappable/
fenceable. Therefore, pageflipping between these 2 FBs where only one
is mappable/fenceable creates latencies large enough to miss alternate
vblanks thereby producing less optimal framerate.
This mainly happens because when i915_gem_object_pin_to_display_plane()
is called to pin one of the FB objs, the associated vma is identified
as misplaced -- because there is no space for it in the aperture --
and therefore i915_vma_unbind() is called which unbinds and evicts it.
This misplaced vma gets subseqently pinned only when
i915_gem_object_ggtt_pin_ww() is called without PIN_MAPPABLE. This whole
thing results in a latency of ~10ms and happens every other repaint cycle.
Therefore, to fix this issue, we just ensure that the misplaced VMA
does not get evicted when we try to pin it with PIN_MAPPABLE -- by
returning early if the mappable/fenceable flag is not set.
Testcase:
Running Weston and weston-simple-egl on an Alderlake_S (ADLS) platform
with a 8K@60 mode results in only ~40 FPS (compared to ~59 FPS with
this patch). Since upstream Weston submits a frame ~7ms before the
next vblank, the latencies seen between atomic commit and flip event
are 7, 24 (7 + 16.66), 7, 24..... suggesting that it misses the
vblank every other frame.
Here is the ftrace snippet that shows the source of the ~10ms latency:
i915_gem_object_pin_to_display_plane() {
0.102 us | i915_gem_object_set_cache_level();
i915_gem_object_ggtt_pin_ww() {
0.390 us | i915_vma_instance();
0.178 us | i915_vma_misplaced();
i915_vma_unbind() {
__i915_active_wait() {
0.082 us | i915_active_acquire_if_busy();
0.475 us | }
intel_runtime_pm_get() {
0.087 us | intel_runtime_pm_acquire();
0.259 us | }
__i915_active_wait() {
0.085 us | i915_active_acquire_if_busy();
0.240 us | }
__i915_vma_evict() {
ggtt_unbind_vma() {
gen8_ggtt_clear_range() {
10507.255 us | }
10507.689 us | }
10508.516 us | }
v2:
- Expand the code comments to describe the ping-pong issue.
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Vivek Kasireddy <vivek.kasireddy@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220321005431.1113890-1-vivek.kasireddy@intel.com
For modern platforms the spec explicitly states that a
SAGV block time of zero means that SAGV is not supported.
Let's extend that to all platforms. Supposedly there should
be no systems where this isn't true, and it'll allow us to:
- use the same code regardless of older vs. newer platform
- wm latencies already treat 0 as disabled, so this fits well
with other related code
- make it a bit more clear when SAGV is used vs. not
- avoid overflows from adding U32_MAX with a u16 wm0 latency value
which could cause us to miscalculate the SAGV watermarks on tgl+
Cc: stable@vger.kernel.org
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220309164948.10671-2-ville.syrjala@linux.intel.com
Reviewed-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com>
(cherry picked from commit d8f5855b31)
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Throttling here refers to the GT frequency being clipped. Each of
the throttle reason attributes will have a 0 or 1 value depending
upon whether there is throttling and also the specific reason for
it.
The following is a brief description of the sysfs throttle
frequency attributes added:
- throttle_reason_status: when set indicates that there is GT
frequency clipping.
- throttle_reason_pl1: when set indicates that PBM PL1 (platform
or package PL1) has caused GT frequency clipping.
- throttle_reason_pl2: when set indicates that PBM PL2 or PL3
(platform or package PL2 or PL3) has caused GT frequency
clipping.
- throttle_reason_pl4: when set indicates that PL4 or IccMax has
caused GT frequency clipping.
- throttle_reason_thermal: when set indicates that Thermal event
has caused GT frequency clipping.
- throttle_reason_prochot: when set indicates that PROCHOT# has
caused GT frequency clipping.
- throttle_reason_ratl: when set indicates that Running Average
Thermal Limit has caused GT frequency clipping.
- throttle_reason_vr_thermalert: when set indicates that Hot VR
(any processor VR) has caused GT frequency clipping.
- throttle_reason_vr_tdc: when set indicates that VR TDC
(Thermal Design Current) has caused GT frequency clipping.
Signed-off-by: Sujaritha Sundaresan <sujaritha.sundaresan@intel.com>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Cc: Dale B Stimson <dale.b.stimson@intel.com>
Reviewed-by: Andrzej Hajda <andrzej.hajda@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220318233938.149744-8-andi.shyti@linux.intel.com
Now tiles have their own sysfs interfaces under the gt/
directory. Because RPS is a property that can be configured on a
tile basis, then each tile should have its own interface
The new sysfs structure will have a similar layout for the 4 tile
case:
/sys/.../card0
├── gt
│ ├── gt0
│ │ ├── id
│ │ ├── rc6_enable
│ │ ├── rc6_residency_ms
│ │ ├── rps_act_freq_mhz
│ │ ├── rps_boost_freq_mhz
│ │ ├── rps_cur_freq_mhz
│ │ ├── rps_max_freq_mhz
│ │ ├── rps_min_freq_mhz
│ │ ├── rps_RP0_freq_mhz
│ │ ├── rps_RP1_freq_mhz
│ │ └── rps_RPn_freq_mhz
. .
. .
. .
│ └── gtN
│ ├── id
│ ├── rc6_enable
│ ├── rc6_residency_ms
│ ├── rps_act_freq_mhz
│ ├── rps_boost_freq_mhz
│ ├── rps_cur_freq_mhz
│ ├── rps_max_freq_mhz
│ ├── rps_min_freq_mhz
│ ├── rps_RP0_freq_mhz
│ ├── rps_RP1_freq_mhz
│ └── rps_RPn_freq_mhz
├── gt_act_freq_mhz -+
├── gt_boost_freq_mhz |
├── gt_cur_freq_mhz | Original interface
├── gt_max_freq_mhz +─-> kept as existing ABI;
├── gt_min_freq_mhz | it points to gt0/
├── gt_RP0_freq_mhz |
├── gt_RP1_freq_mhz |
└── gt_RPn_freq_mhz -+
The existing interfaces have been kept in their original location
to preserve the existing ABI. They act on all the GTs: when
writing they loop through all the GTs and write the information
on each interface. When reading they provide the average value
from all the GTs.
This patch is not really adding exposing new interfaces (new
ABI) other than adapting the existing one to more tiles. In any
case this new set of interfaces will be a basic tool for system
managers and administrators when using i915.
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Cc: Sujaritha Sundaresan <sujaritha.sundaresan@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Andrzej Hajda <andrzej.hajda@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220318233938.149744-7-andi.shyti@linux.intel.com
Now tiles have their own sysfs interfaces under the gt/
directory. Because RC6 is a property that can be configured on a
tile basis, then each tile should have its own interface
The new sysfs structure will have a similar layout for the 4 tile
case:
/sys/.../card0
├── gt
│ ├── gt0
│ │ ├── id
│ │ ├── rc6_enable
│ │ ├── rc6_residency_ms
. . .
. . .
. .
│ └── gtN
│ ├── id
│ ├── rc6_enable
│ ├── rc6_residency_ms
│ .
│ .
│
└── power/ -+
├── rc6_enable | Original interface
├── rc6_residency_ms +-> kept as existing ABI;
. | it multiplexes over
. | the GTs
-+
The existing interfaces have been kept in their original location
to preserve the existing ABI. They act on all the GTs: when
reading they provide the average value from all the GTs.
This patch is not really adding exposing new interfaces (new
ABI) other than adapting the existing one to more tiles. In any
case this new set of interfaces will be a basic tool for system
managers and administrators when using i915.
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Cc: Sujaritha Sundaresan <sujaritha.sundaresan@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Andrzej Hajda <andrzej.hajda@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220318233938.149744-6-andi.shyti@linux.intel.com
Add logical mapping for VDBOXs. This mapping is required for
split-frame workloads, which otherwise fail with
00000000-F8C53528: [GUC] 0441-INVALID_ENGINE_SUBMIT_MASK
... if the application is using the logical id to reorder the engines and
then using it for the batch buffer submission. It's not a big problem on
media version 11 and 12 as they have only 2 instances of VCS and the
logical to physical mapping is monotonically increasing - if the
application is not using the logical id.
Changing it for the previous platforms allows the media driver
implementation for the next ones (12.50 and above) to be the same,
checking the logical id. It should also not introduce any bug for the
old versions of userspace not checking the id.
The mapping added here is the complete map needed by XEHPSDV. Previous
platforms with only 2 instances will just use a partial map and should
still work.
v2: Remove static from map variable (José)
Cc: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
[ Extend the mapping to media versions 11 and 12 and give proper
justification in the commit message why ]
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Acked-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220316234538.434357-2-lucas.demarchi@intel.com
For modern platforms the spec explicitly states that a
SAGV block time of zero means that SAGV is not supported.
Let's extend that to all platforms. Supposedly there should
be no systems where this isn't true, and it'll allow us to:
- use the same code regardless of older vs. newer platform
- wm latencies already treat 0 as disabled, so this fits well
with other related code
- make it a bit more clear when SAGV is used vs. not
- avoid overflows from adding U32_MAX with a u16 wm0 latency value
which could cause us to miscalculate the SAGV watermarks on tgl+
Cc: stable@vger.kernel.org
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220309164948.10671-2-ville.syrjala@linux.intel.com
Reviewed-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com>
Implement support for fetching the hardware description table from the
GuC. The call is made twice - once without a destination buffer to
query the size and then a second time to fill in the buffer.
The table is stored in the GT structure so that it can be fetched once
at driver load time. Keeping inside a GuC structure would mean it
would be release and reloaded on a GuC reset (part of a full GT
reset). However, the table does not change just because the GT has been
reset and the GuC reloaded. Also, dynamic memory allocations inside
the reset path are a problem.
Note that the table is only available on ADL-P and later platforms.
v2 (John's v2 patch):
* Move to GT level to avoid memory allocation during reset path (and
unnecessary re-read of the table on a reset).
v5 (of Jordan's posting):
* Various changes made by Jordan and recommended by Michal
- Makefile ordering
- Adjust "struct intel_guc_hwconfig hwconfig" comment
- Set Copyright year to 2022 in intel_guc_hwconfig.c/.h
- Drop inline from hwconfig_to_guc()
- Replace hwconfig param with guc in __guc_action_get_hwconfig()
- Move zero size check into guc_hwconfig_discover_size()
- Change comment to say zero size offset/size is needed to get size
- Add has_guc_hwconfig to devinfo and drop has_table()
- Change drm_err to notice in __uc_init_hw() and use %pe
v6 (of Jordan's posting):
* Added a couple more small changes recommended by Michal
* Merge in John's v2 patch, but note:
- Using drm_notice as recommended by Michal
- Reverted Michal's suggestion of using devinfo
v7 (of Jordan's posting):
* Change back to drm_err as preferred by John
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Acked-by: Jon Bloomfield <jon.bloomfield@intel.com>
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220306232157.1174335-2-jordan.l.justen@intel.com
On integrated it looks like the GGTT base should always 1:1 maps to
somewhere within DSM. On discrete the base seems to be pre-programmed with
a normal lmem address, and is not 1:1 mapped with the base address. On
such devices probe the lmem address directly from the PTE.
v2(Ville):
- The base is actually the pre-programmed GGTT address, which is then
meant to 1:1 map to somewhere inside dsm. In the case of dgpu the
base looks to just be some offset within lmem, but this also happens
to be the exact dsm start, on dg1. Therefore we should only need to
fudge the physical address, before allocating from stolen.
- Bail if it's not located in dsm.
v3:
- Scratch that. There doesn't seem to be any relationship with the
base and PTE address, on at least DG1. Let's instead just grab the
lmem address from the PTE itself.
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Nirmoy Das <nirmoy.das@linux.intel.com>
Reviewed-by: Nirmoy Das <nirmoy.das@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220315181425.576828-7-matthew.auld@intel.com
