This test will validate we can achieve actual frequency of RP0. Pcode
grants frequencies based on what GuC is requesting. However, thermal
throttling can limit what is being granted. Add a test to request for
max, but don't fail the test if RP0 is not granted due to throttle
reasons.
Also optimize the selftest by using a common run_test function to avoid
code duplication. Rename the "clamp" tests to vary_max_freq and
vary_min_freq.
v2: Fix compile warning
v3: Review comments (Ashutosh). Added a FIXME for the media RP0 case.
v4: Checkpatch (strict) fixes, remove FIXME and other comments (Ashutosh)
Fixes commit 8ee2c22782 ("drm/i915/guc/slpc: Add SLPC selftest")
Cc: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220627230346.27720-1-vinay.belgaumkar@intel.com
SLPC min/max frequency updates require H2G calls. We are seeing
timeouts when GuC channel is backed up and it is unable to respond
in a timely fashion causing warnings and affecting CI.
This is seen when waitboosting happens during a stress test.
this patch updates the waitboost path to use a non-blocking
H2G call instead, which returns as soon as the message is
successfully transmitted.
v2: Use drm_notice to report any errors that might occur while
sending the waitboost H2G request (Tvrtko)
v3: Add drm_notice inside force_min_freq (Ashutosh)
Cc: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220623003225.23301-1-vinay.belgaumkar@intel.com
For execlists backend, current implementation of Wa_22011802037 is to
stop the CS before doing a reset of the engine. This WA was further
extended to wait for any pending MI FORCE WAKEUPs before issuing a
reset. Add the extended steps in the execlist path of reset.
In addition, extend the WA to gen11.
v2: (Tvrtko)
- Clarify comments, commit message, fix typos
- Use IS_GRAPHICS_VER for gen 11/12 checks
v3: (Daneile)
- Drop changes to intel_ring_submission since WA does not apply to it
- Log an error if MSG IDLE is not defined for an engine
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Fixes: f6aa0d713c ("drm/i915: Add Wa_22011802037 force cs halt")
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220621192105.2100585-1-umesh.nerlige.ramappa@intel.com
Using two different types of workoads, it was observed that
guc_update_engine_gt_clks was being called too frequently and/or
causing a CPU-to-lmem bandwidth hit over PCIE. Details on
the workloads and numbers are in the notes below.
Background: At the moment, guc_update_engine_gt_clks can be invoked
via one of 3 ways. #1 and #2 are infrequent under normal operating
conditions:
1.When a predefined "ping_delay" timer expires so that GuC-
busyness can sample the GTPM clock counter to ensure it
doesn't miss a wrap-around of the 32-bits of the HW counter.
(The ping_delay is calculated based on 1/8th the time taken
for the counter go from 0x0 to 0xffffffff based on the
GT frequency. This comes to about once every 28 seconds at a
GT frequency of 19.2Mhz).
2.In preparation for a gt reset.
3.In response to __gt_park events (as the gt power management
puts the gt into a lower power state when there is no work
being done).
Root-cause: For both the workloads described farther below, it was
observed that when user space calls IOCTLs that unparks the
gt momentarily and repeats such calls many times in quick succession,
it triggers calling guc_update_engine_gt_clks as many times. However,
the primary purpose of guc_update_engine_gt_clks is to ensure we don't
miss the wraparound while the counter is ticking. Thus, the solution
is to ensure we skip that check if gt_park is calling this function
earlier than necessary.
Solution: Snapshot jiffies when we do actually update the busyness
stats. Then get the new jiffies every time intel_guc_busyness_park
is called and bail if we are being called too soon. Use half of the
ping_delay as a safe threshold.
NOTE1: Workload1: IGTs' gem_create was modified to create a file handle,
allocate memory with sizes that range from a min of 4K to the max supported
(in power of two step-sizes). Its maps, modifies and reads back the
memory. Allocations and modification is repeated until total memory
allocation reaches the max. Then the file handle is closed. With this
workload, guc_update_engine_gt_clks was called over 188 thousand times
in the span of 15 seconds while this test ran three times. With this patch,
the number of calls reduced to 14.
NOTE2: Workload2: 30 transcode sessions are created in quick succession.
While these sessions are created, pcm-iio tool was used to measure I/O
read operation bandwidth consumption sampled at 100 milisecond intervals
over the course of 20 seconds. The total bandwidth consumed over 20 seconds
without this patch was measured at average at 311KBps per sample. With this
patch, the number went down to about 175Kbps which is about a 43% savings.
Signed-off-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220623023157.211650-2-alan.previn.teres.alexis@intel.com
This reverts commit 1e98d8c52e.
The problem with this patch is that it makes i915_request to hold a
reference to intel_context, which in turn holds a reference on the VM.
This strong back referencing can lead to reference loops which leads
to resource leak.
An example is the upcoming VM_BIND work which requires VM to hold
a reference to some shared VM specific BO. But this BO's dma-resv
fences holds reference to the i915_request thus leading to reference
loop.
v2:
Do not use reserved requests for virtual engines
Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Ramalingam C <ramalingam.c@intel.com>
Suggested-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Mathew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220614184348.23746-3-ramalingam.c@intel.com
Re-do what was attempted in commit 7a5c922377 ("drm/i915/gt: Split
intel-gtt functions by arch"). The goal of that commit was to split the
handlers for older hardware that depend on intel-gtt.ko so i915 can
be built for non-x86 archs, after some more patches. Other archs do not
need intel-gtt.ko.
Main issue with the previous approach: it moved all the hooks, including
the gen8, which is used by all platforms gen8 and newer. Re-do the
split moving only the handlers for gen < 6, which are the only ones
calling out to the separate module.
While at it do some minor cleanups:
- Rename the prefix s/gen5_/gmch_/ to be more accurate what platforms
are covered by intel_ggtt_gmch.c
- Remove dead code for gen12 out of needs_idle_maps()
- Remove TODO comment leftover
- Re-order if/else ladder in ggtt_probe_hw() to keep newest platforms
first
v2: Add minor cleanups (Matt Roper)
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220617230559.2109427-2-lucas.demarchi@intel.com
When DMAR / VT-d is enabled, the display engine uses overfetching,
presumably to deal with the increased latency. To avoid display engine
errors and DMAR faults, as a workaround the GGTT is populated with scatch
PTEs when VT-d is enabled. However starting with gen10, Write-combined
writing of scratch PTES is no longer possible and as a result, populating
the full GGTT with scratch PTEs like on resume becomes very slow as
uncached access is needed.
Therefore, on integrated GPUs utilize the fact that the PTEs are stored in
stolen memory which retain content across S3 suspend. Don't clear the PTEs
on suspend and resume. This improves on resume time with around 100 ms.
While 100+ms might appear like a short time it's 10% to 20% of total resume
time and important in some applications.
One notable exception is Intel Rapid Start Technology which may cause
stolen memory to be lost across what the OS percieves as S3 suspend.
If IRST is enabled or if we can't detect whether IRST is enabled, retain
the old workaround, clearing and re-instating PTEs.
As an additional measure, if we detect that the last ggtt pte was lost
during suspend, print a warning and re-populate the GGTT ptes
On discrete GPUs, the display engine scans out from LMEM which isn't
subject to DMAR, and presumably the workaround is therefore not needed,
but that needs to be verified and disabling the workaround for dGPU,
if possible, will be deferred to a follow-up patch.
v2:
- Rely on retained ptes to also speed up suspend and resume re-binding.
- Re-build GGTT ptes if Intel rst is enabled.
v3:
- Re-build GGTT ptes also if we can't detect whether Intel rst is enabled,
and if the guard page PTE and end of GGTT was lost.
v4:
- Fix some kerneldoc issues (Matthew Auld), rebase.
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220617152856.249295-1-thomas.hellstrom@linux.intel.com
Let's replace the assortment of intel_gt_* and intel_uncore_* functions
that operate on MCR registers with a cleaner set of interfaces:
* intel_gt_mcr_read -- unicast read from specific instance
* intel_gt_mcr_read_any[_fw] -- unicast read from any non-terminated
instance
* intel_gt_mcr_unicast_write -- unicast write to specific instance
* intel_gt_mcr_multicast_write[_fw] -- multicast write to all instances
We'll also replace the historic "slice" and "subslice" terminology with
"group" and "instance" to match the documentation for more recent
platforms; these days MCR steering applies to more types of replication
than just slice/subslice.
v2:
- Reference the new kerneldoc from i915.rst. (Jani)
- Tweak the wording of the documentation for a couple functions to
clarify the difference between "_fw" and non-"_fw" forms.
v3:
- s/read/write/ to fix copy-paste mistake in a couple comments.
(Harish)
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Acked-by: Jani Nikula <jani.nikula@linux.intel.com>
Reviewed-by: Harish Chegondi <harish.chegondi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220615001019.1821989-3-matthew.d.roper@intel.com
Handling of multicast/replicated registers is spread across intel_gt.c
and intel_uncore.c today. As multicast handling and the related
steering logic gets more complicated with the addition of new platforms
and new rules it makes sense to centralize it all in one place.
For now the existing functions have been moved to the new .c/.h as-is.
Function renames and updates to operate in a more consistent manner will
be done in subsequent patches.
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Acked-by: Jani Nikula <jani.nikula@linux.intel.com>
Reviewed-by: Harish Chegondi <harish.chegondi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220615001019.1821989-2-matthew.d.roper@intel.com
We have long standing customer complaints that pressing Ctrl-C (or to the
effect of) causes engine resets with otherwise well behaving programs.
Not only is logging engine resets during normal operation not desirable
since it creates support incidents, but more fundamentally we should avoid
going the engine reset path when we can since any engine reset introduces
a chance of harming an innocent context.
Reason for this undesirable behaviour is that the driver currently does
not distinguish between banned contexts and non-persistent contexts which
have been closed.
To fix this we add the distinction between the two reasons for revoking
contexts, which then allows the strict timeout only be applied to banned,
while innocent contexts (well behaving) can preempt cleanly and exit
without triggering the engine reset path.
Note that the added context exiting category applies both to closed non-
persistent context, and any exiting context when hangcheck has been
disabled by the user.
At the same time we rename the backend operation from 'ban' to 'revoke'
which more accurately describes the actual semantics. (There is no ban at
the backend level since banning is a concept driven by the scheduling
frontend. Backends are simply able to revoke a running context so that
is the more appropriate name chosen.)
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Andrzej Hajda <andrzej.hajda@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220527072452.2225610-1-tvrtko.ursulin@linux.intel.com
Ponte Vecchio no longer has MSLICE or LNCF steering, but the bspec does
document several new types of multicast register ranges. Fortunately,
most of the different MCR types all provide valid values at instance
(0,0) so there's no need to read fuse registers and calculate a
non-terminated instance. We'll lump all of those range types (BSLICE,
HALFBSLICE, TILEPSMI, CC, and L3BANK) into a single category called
"INSTANCE0" to keep things simple. We'll also perform explicit steering
for each of these multicast register types, even if the implicit
steering setup for COMPUTE/DSS ranges would have worked too; this is
based on guidance from our hardware architects who suggested that we
move away from implicit steering and start explicitly steer all MCR
register accesses on modern platforms (we'll work on transitioning
COMPUTE/DSS to explicit steering in the future).
Note that there's one additional MCR range type defined in the bspec
(SQIDI) that we don't handle here. Those ranges use a different
steering control register that we never touch; since instance 0 is also
always a valid setting there, we can just ignore those ranges.
Finally, we'll rename the HAS_MSLICES() macro to HAS_MSLICE_STEERING().
PVC hardware still has units referred to as mslices, but there's no
register steering based on mslice for this platform.
v2:
- Rebase on other recent changes
- Swap two table rows to keep table sorted & easy to read. (Harish)
Bspec: 67609
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Harish Chegondi <harish.chegondi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220608170700.4026648-1-matthew.d.roper@intel.com
A new PVC+DG2 workaround has appeared recently:
- Wa_16015675438
And a couple existing DG2 workarounds have been extended to PVC:
- Wa_14015795083
- Wa_18018781329
Note that Wa_16015675438 asks us to program a register that is in the
0x2xxx range typically associated with the RCS engine, even though PVC
does not have an RCS. By default the GuC will think we've made a
mistake and throw an exception when it sees this register on a CCS
engine's save/restore list, so we need to pass an extra GuC control flag
to tell it that this is expected and not a problem.
Signed-off-by: Anshuman Gupta <anshuman.gupta@intel.com>
Signed-off-by: Badal Nilawar <badal.nilawar@intel.com>
Signed-off-by: Prathap Kumar Valsan <prathap.kumar.valsan@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Anshuman Gupta <anshuman.gupta@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220608005108.3717895-1-matthew.d.roper@intel.com
Our internal teams have identified a few additional engine registers
that are worth inspecting in error state dumps during development &
debug. Let's capture and print them as part of our error dump.
For simplicity we'll just dump these registers on gen11 and beyond.
Most of these registers have existed since earlier platforms (e.g., gen6
or gen7) but were initially introduced only for a subset of the
platforms' engines; gen11 seems to be where they became available on all
engines.
Signed-off-by: Stuart Summers <stuart.summers@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220601210646.615946-1-matthew.d.roper@intel.com
As with EU masks, it's easier to store subslice/DSS masks internally in
a format that's more natural for the driver to work with, and then only
covert into the u8[] uapi form when the query ioctl is invoked. Since
the hardware design changed significantly with Xe_HP, we'll use a union
to choose between the old "hsw-style" subslice masks or the newer xehp
mask. HSW-style masks will be stored in an array of u8's, indexed by
slice (there's never more than 6 subslices per slice on older
platforms). For Xe_HP and beyond where slices no longer exist, we only
need a single bitmask. However we already know that this mask is
eventually going to grow too large for a simple u64 to hold, so we'll
represent it in a manner that can be operated on by the utilities in
linux/bitmap.h.
v2:
- Fix typo: BIT(s) -> BIT(ss) in gen9_sseu_device_status()
v3:
- Eliminate sseu->ss_stride and just calculate the stride while
specifically handling uapi. (Tvrtko)
- Use BITMAP_BITS() macro to refer to size of masks rather than
passing I915_MAX_SS_FUSE_BITS directly. (Tvrtko)
- Report compute/geometry DSS masks separately when dumping Xe_HP SSEU
info. (Tvrtko)
- Restore dropped range checks to intel_sseu_has_subslice(). (Tvrtko)
v4:
- Make the bitmap size macro check the size of the .xehp field rather
than the containing union. (Tvrtko)
- Don't add GEM_BUG_ON() intel_sseu_has_subslice()'s check for whether
slice or subslice ID exceed sseu->max_[sub]slices; various loops
in the driver are expected to exceed these, so we should just
silently return 'false.'
v5:
- Move XEHP_BITMAP_BITS() to the header so that we can also replace a
usage of I915_MAX_SS_FUSE_BITS in one of the inline functions.
(Bala)
- Change the local variable in intel_slicemask_from_xehp_dssmask() from
u16 to 'unsigned long' to make it a bit more future-proof.
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Cc: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220601150725.521468-6-matthew.d.roper@intel.com
Storing the EU mask internally in the same format the I915_QUERY
topology queries use makes the final copy_to_user() a bit simpler, but
makes the rest of the driver's SSEU more complicated and harder to
follow. Let's switch to an internal representation that's more natural:
Xe_HP platforms will be a simple array of u16 masks, whereas pre-Xe_HP
platforms will be a two-dimensional array, indexed by [slice][subslice].
We'll convert to the uapi format only when the query uapi is called.
v2:
- Drop has_common_ss_eumask. We waste some space repeating identical
EU masks for every single DSS, but the code is simpler without it.
(Tvrtko)
v3:
- Mask down EUs passed to sseu_set_eus at the callsite rather than
inside the function. (Tvrtko)
- Eliminate sseu->eu_stride and calculate it when needed. (Tvrtko)
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220601150725.521468-5-matthew.d.roper@intel.com
Slice/subslice/EU information should be obtained via the topology
queries provided by the I915_QUERY interface; let's turn off support for
the old GETPARAM lookups on Xe_HP and beyond where we can't return
meaningful values.
The slice mask lookup is meaningless since Xe_HP doesn't support
traditional slices (and we make no attempt to return the various new
units like gslices, cslices, mslices, etc.) here.
The subslice mask lookup is even more problematic; given the distinct
masks for geometry vs compute purposes, the combined mask returned here
is likely not what userspace would want to act upon anyway. The value
is also limited to 32-bits by the nature of the GETPARAM ioctl which is
sufficient for the initial Xe_HP platforms, but is unable to convey the
larger masks that will be needed on other upcoming platforms. Finally,
the value returned here becomes even less meaningful when used on
multi-tile platforms where each tile will have its own masks.
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com> # mesa
Reviewed-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220601150725.521468-3-matthew.d.roper@intel.com
For PVC, the base die and compute tile have separate stepping values
that we need to track; we'll use the existing graphics_step field to
represent the compute tile stepping and add a new 'basedie_step' field.
Unlike past platforms, steppings for these components are represented by
specific bitfields within the PCI revision ID, and we shouldn't make
assumptions about the non-CT, non-BD bits staying 0. Let's update our
stepping code accordingly.
Bspec: 44484
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Matt Atwood <matthew.s.atwood@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220527163348.1936146-2-matthew.d.roper@intel.com
_i915_vma_move_to_active() can receive > 1 fences for
multiple batch buffers submission. Because dma_resv_add_fence()
can only accept one fence at a time, change _i915_vma_move_to_active()
to be aware of multiple fences so that it can add individual
fences to the dma resv object.
v6: fix multi-line comment.
v5: remove double fence reservation for batch VMAs.
v4: Reserve fences for composite_fence on multi-batch contexts and
also reserve fence slots to composite_fence for each VMAs.
v3: dma_resv_reserve_fences is not cumulative so pass num_fences.
v2: make sure to reserve enough fence slots before adding.
Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/5614
Fixes: 544460c338 ("drm/i915: Multi-BB execbuf")
Cc: <stable@vger.kernel.org> # v5.16+
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@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/20220525095955.15371-1-nirmoy.das@intel.com
