mei: mkhi: add memory ready command

Add GSC memory ready command.
The command indicates to the firmware that extend operation
memory was setup and the firmware may enter PXP mode.

CC: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220907215113.1596567-11-tomas.winkler@intel.com
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

mei: bus: export common mkhi definitions into a separate header

Exported common mkhi definitions from bus-fixup.c into a separate
header file mkhi.h for other driver usage.

Signed-off-by: Vitaly Lubart <vitaly.lubart@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220907215113.1596567-10-tomas.winkler@intel.com
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

mei: extend timeouts on slow devices

Parametrize operational timeouts in order
to support slow firmware on some graphics devices.

Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220907215113.1596567-9-tomas.winkler@intel.com
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

mei: gsc: wait for reset thread on stop

Wait for reset work to complete before initiating
stop reset flow sequence.

Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220907215113.1596567-8-tomas.winkler@intel.com
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

mei: gsc: use polling instead of interrupts

A work-around for a HW issue in XEHPSDV that manifests itself when SW reads
a gsc register when gsc is sending an interrupt. The work-around is
to disable interrupts and to use polling instead.

Cc: James Ausmus <james.ausmus@intel.com>
Signed-off-by: Vitaly Lubart <vitaly.lubart@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220907215113.1596567-7-tomas.winkler@intel.com
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

drm/i915/gsc: add GSC XeHP SDV platform definition

Define GSC on XeHP SDV (Intel(R) dGPU without display)

XeHP SDV uses the same hardware settings as DG1, but uses polling
instead of interrupts and runs the firmware in slow pace due to
hardware limitations.

Signed-off-by: Vitaly Lubart <vitaly.lubart@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220907215113.1596567-6-tomas.winkler@intel.com
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

drm/i915/gsc: add slow_firmware flag to the gsc device definition

Add slow_firmware flag to the gsc device definition
and pass it to mei auxiliary device, this instructs
the driver to use longer operation timeouts.

Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220907215113.1596567-5-tomas.winkler@intel.com
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

mei: add slow_firmware flag to the mei auxiliary device

Add slow_firmware flag to the mei auxiliary device info
to inform the mei driver about slow underlying firmware.
Such firmware will require to use larger operation timeouts.

Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220907215113.1596567-4-tomas.winkler@intel.com
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

mei: add kdoc for struct mei_aux_device

struct mei_aux_device is an interface structure
requires proper documenation.

Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220907215113.1596567-3-tomas.winkler@intel.com
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

drm/i915/gsc: skip irq initialization if using polling

Some platforms require the host to poll on the
GSC registers instead of relaying on the interrupts.
For those platforms, irq initialization should be skipped

Signed-off-by: Vitaly Lubart <vitaly.lubart@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220907215113.1596567-2-tomas.winkler@intel.com
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

drm/i915: Set correct domains values at _i915_vma_move_to_active

Fix regression introduced by commit:
"drm/i915: Individualize fences before adding to dma_resv obj"
which sets obj->read_domains to 0 for both read and write paths.
Also set obj->write_domain to 0 on read path which was removed by
the commit.

References: https://gitlab.freedesktop.org/drm/intel/-/issues/6639
Fixes: 420a07b841d0 ("drm/i915: Individualize fences before adding to dma_resv obj")
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
Cc: <stable@vger.kernel.org> # v5.16+
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Andrzej Hajda <andrzej.hajda@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/20220907172641.12555-1-nirmoy.das@intel.com

drm/i915: Rename ggtt_view as gtt_view

So far, different views (normal, partial, rotated and remapped)
into the same object are only supported for GGTT mappings.
But with the upcoming VM_BIND feature, PPGTT will also use the
partial view mapping. Hence rename ggtt_view to more generic
gtt_view.

Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220901183854.3446-1-niranjana.vishwanathapura@intel.com

drm/i915/uc: Add patch level version number support

With the move to un-versioned filenames, it becomes more difficult to
know exactly what version of a given firmware is being used. So add
the patch level version number to the debugfs output.

Also, support matching by patch level when selecting code paths for
firmware compatibility. While a patch level change cannot be backwards
breaking, it is potentially possible that a new feature only works
from a given patch level onwards (even though it was theoretically
added in an earlier version that bumped the major or minor version).

Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220906230147.479945-2-daniele.ceraolospurio@intel.com

drm/i915/uc: Support for version reduced and multiple firmware files

There was a misunderstanding in how firmware file compatibility should
be managed within i915. This has been clarified as:
  i915 must support all existing firmware releases forever
  new minor firmware releases should replace prior versions
  only backwards compatibility breaking releases should be a new file

This patch cleans up the single fallback file support that was added
as a quick fix emergency effort. That is now removed in preference to
supporting arbitrary numbers of firmware files per platform.

The patch also adds support for having GuC firmware files that are
named by major version only (because the major version indicates
backwards breaking changes that affect the KMD) and for having HuC
firmware files with no version number at all (because the KMD has no
interface requirements with the HuC).

For GuC, the driver will report via dmesg if the found file is older than
expected. For HuC, the KMD will no longer require updating for any new
HuC release so will not be able to report what the latest expected
version is.

Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220906230147.479945-1-daniele.ceraolospurio@intel.com

drm/i915: Don't try to disable host RPS when this was never enabled.

Specially in GT reset case this could be triggered and try
to disable things that had never been enabled. Let's add
some protection here.

Cc: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Acked-by: Nirmoy Das <nirmoy.das@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220902095126.373036-1-rodrigo.vivi@intel.com

drm/i915: consider HAS_FLAT_CCS() in needs_ccs_pages

Just move the HAS_FLAT_CCS() check into needs_ccs_pages. This also then
fixes i915_ttm_memcpy_allowed() which was incorrectly reporting true on
DG1, even though it doesn't have small-BAR or flat-CCS.

References: https://gitlab.freedesktop.org/drm/intel/-/issues/6605
Fixes: efeb3caf4341 ("drm/i915/ttm: disallow CPU fallback mode for ccs pages")
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220905105329.41455-1-matthew.auld@intel.com

drm/i915/ttm: Abort suspend on i915_ttm_backup failure

On system suspend when system memory is low then i915_gem_obj_copy_ttm()
could fail trying to backup a lmem obj. GEM_WARN_ON() is not enough,
suspend shouldn't continue if i915_ttm_backup() throws an error.

v2: Keep the fdo issue till we have a igt test(Matt).
v3: Use %pe(Andrzej)

References: https://gitlab.freedesktop.org/drm/intel/-/issues/6529
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Andrzej Hajda <andrzej.hajda@intel.com>
Suggested-by: Chris P Wilson <chris.p.wilson@intel.com>
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220901172217.18392-1-nirmoy.das@intel.com

drm/i915/slpc: Let's fix the PCODE min freq table setup for SLPC

We need to inform PCODE of a desired ring frequencies so PCODE update
the memory frequencies to us. rps->min_freq and rps->max_freq are the
frequencies used in that request. However they were unset when SLPC was
enabled and PCODE never updated the memory freq.

v2 (as Suggested by Ashutosh): if SLPC is in use, let's pick the right
   frequencies from the get_ia_constants instead of the fake init of
   rps' min and max.

v3: don't forget the max <= min return

v4: Move all the freq conversion to intel_rps.c. And the max <= min
    check to where it belongs.

v5: (Ashutosh) Fix old comment s/50 HZ/50 MHz and add a doc explaining
    the "raw format"

Fixes: 7ba79a671568 ("drm/i915/guc/slpc: Gate Host RPS when SLPC is enabled")
Cc: <stable@vger.kernel.org> # v5.15+
Cc: Ashutosh Dixit <ashutosh.dixit@intel.com>
Tested-by: Sushma Venkatesh Reddy <sushma.venkatesh.reddy@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220831214538.143950-1-rodrigo.vivi@intel.com

drm/i915/slpc: Fix inconsistent locked return

Fix for intel_guc_slpc_set_min_freq() warn:
inconsistent returns '&slpc->lock'.

v2: Avoid with_intel_runtime_pm with the
    internal goto/return. (Ashutosh)
    Also standardize the 'ret' if this came from
    the efficient setup. And avoid the 'unlikely'.

Fixes: 95ccf312a1e4 ("drm/i915/guc/slpc: Allow SLPC to use efficient frequency")
Reported-by: kernel test robot <lkp@intel.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220830193537.52201-1-rodrigo.vivi@intel.com

drm/i915/ats-m: Add thread execution tuning setting

On client DG2 platforms, optimal performance is achieved with the
hardware's default "age based" thread execution setting.  However on
ATS-M, switching this to "round robin after dependencies" provides
better performance.  We'll add a new "tuning" feature flag to the ATS-M
device info to enable/disable this setting.

Bspec: 68331
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
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/20220826212718.409948-1-matthew.d.roper@intel.com

Revert "drm/i915/dg2: Add preemption changes for Wa_14015141709"

This reverts commit ca6920811aa5428270dd78af0a7a36b10119065a.

The intent of Wa_14015141709 was to inform us that userspace can no
longer control object-level preemption as it has on past platforms
(i.e., by twiddling register bit CS_CHICKEN1[0]).  The description of
the workaround in the spec wasn't terribly well-written, and when we
requested clarification from the hardware teams we were told that on the
kernel side we should also probably stop setting
FF_SLICE_CS_CHICKEN1[14], which is the register bit that directs the
hardware to honor the settings in per-context register CS_CHICKEN1.  It
turns out that this guidance about FF_SLICE_CS_CHICKEN1[14] was a
mistake; even though CS_CHICKEN1[0] is non-operational and useless to
userspace, there are other bits in the register that do still work and
might need to be adjusted by userspace in the future (e.g., to implement
other workarounds that show up).  If we don't set
FF_SLICE_CS_CHICKEN1[14] in i915, then those future workarounds would
not take effect.

This miscommunication came to light because another workaround
(Wa_16013994831) has now shown up that requires userspace to adjust the
value of CS_CHICKEN[10] in certain circumstances.  To ensure userspace's
updates to this chicken bit are handled properly by the hardware, we
need to make sure that FF_SLICE_CS_CHICKEN1[14] is once again set by the
kernel.

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/20220826210233.406482-1-matthew.d.roper@intel.com

drm/i915/selftests: allow misaligned_pin test work with unmappable memory

In case of Small BAR configurations stolen local memory can be unmappable.
Since the test does not touch the memory, passing I915_BO_ALLOC_GPU_ONLY
flag to i915_gem_object_create_region, will prevent -ENOSPC error from
_i915_gem_object_stolen_init.

Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/6565
Signed-off-by: Andrzej Hajda <andrzej.hajda@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220825154239.52343-1-andrzej.hajda@intel.com

drm/i915/guc: Remove log size module parameters

Remove the module parameters for configuring GuC log size.

We should instead rely on tuning the defaults to be usable for
reporting bugs.

v2:
- Use correct 1M unit

Fixes: 8ad0152afb1b ("drm/i915/guc: Make GuC log sizes runtime configurable")
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Cc: John Harrison <John.C.Harrison@Intel.com>
Cc: Alan Previn <alan.previn.teres.alexis@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220826092343.184568-1-joonas.lahtinen@linux.intel.com

drm/i915/dg2: Incorporate Wa_16014892111 into DRAW_WATERMARK tuning

Although register tuning settings are generally implemented via the
workaround infrastructure, it turns out that the DRAW_WATERMARK register
is not properly saved/restored by hardware around power events (i.e.,
RC6 entry) so updates to the value cannot be applied in the usual
manner.  New workaround Wa_16014892111 informs us that any tuning
updates to this register must instead be applied via an INDIRECT_CTX
batch buffer.  This will ensure that the necessary value is re-applied
when a context begins running, even if an RC6 entry had wiped the
register back to hardware defaults since the last context ran.

Fixes: 6dc85721df74 ("drm/i915/dg2: Add additional tuning settings")
Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/6642
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220823202449.83727-1-matthew.d.roper@intel.com

drm/i915/pxp: don't start pxp without mei_pxp bind

pxp will not start correctly until after mei_pxp bind completes and
intel_pxp_init_hw() is called.
Wait for the bind to complete before proceeding with startup.

This fixes a race condition during bootup where we observed a small
window for pxp commands to be sent, starting pxp before mei_pxp bind
completed.

Changes since v2:
- wait for pxp_component to bind instead of returning -EAGAIN (Daniele)

Changes since v1:
- check pxp_component instead of pxp_component_added (Daniele)
- pxp_component needs tee_mutex (Daniele)
- return -EAGAIN so caller knows to retry (Daniele)

Signed-off-by: Juston Li <justonli@chromium.org>
Reviewed-by: Andrzej Hajda <andrzej.hajda@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220818174205.2412730-1-justonli@chromium.org

drm/i915/mtl: Don't mask off CCS according to DSS fusing

Unlike the Xe_HP platforms, MTL only has a single CCS engine; the
quad-based engine masking logic does not apply to this platform (or
presumably any future platforms that only have 0 or 1 CCS).

Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Radhakrishna Sripada <radhakrishna.sripada@intel.com>
Reviewed-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220818234202.451742-5-radhakrishna.sripada@intel.com

drm/i915/mtl: MMIO range is now 4MB

Previously only dgfx platforms had a 4MB MMIO range, but starting with
MTL we now use the larger range for all platforms.

Bspec: 63834, 63830
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Radhakrishna Sripada <radhakrishna.sripada@intel.com>
Reviewed-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220818234202.451742-4-radhakrishna.sripada@intel.com

drm/i915: Skip Bit12 fw domain reset for gen12+

Bit12 of the Forcewake request register should not be cleared post
gen12. Do not touch this bit while clearing during fw domain reset.

v2: Tweak the comment to drop older platforms(MattR)

Bspec: 52542

Signed-off-by: Sushma Venkatesh Reddy <sushma.venkatesh.reddy@intel.com>
Signed-off-by: Radhakrishna Sripada <radhakrishna.sripada@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220817224304.255767-1-radhakrishna.sripada@intel.com

drm/i915/guc/slpc: Allow SLPC to use efficient frequency

Host Turbo operates at efficient frequency when GT is not idle unless
the user or workload has forced it to a higher level. Replicate the same
behavior in SLPC by allowing the algorithm to use efficient frequency.
We had disabled it during boot due to concerns that it might break
kernel ABI for min frequency. However, this is not the case since
SLPC will still abide by the (min,max) range limits.

With this change, min freq will be at efficient frequency level at init
instead of fused min (RPn). If user chooses to reduce min freq below the
efficient freq, we will turn off usage of efficient frequency and honor
the user request. When a higher value is written, it will get toggled
back again.

The patch also corrects the register which needs to be read for obtaining
the correct efficient frequency for Gen9+.

We see much better perf numbers with benchmarks like glmark2 with
efficient frequency usage enabled as expected.

v2: Address review comments (Rodrigo)
v3: with efficient frequency being dynamic, it is possible that the req
frequency may go beyond max freq. This will cause SLPC selftests to fail.
Add a FIXME there to start the test with [RPn, RP0] instead and restore
it afterwards.

BugLink: https://gitlab.freedesktop.org/drm/intel/-/issues/5468
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220820010832.15350-1-vinay.belgaumkar@intel.com

drm/i915/guc: remove runtime info printing from time stamp logging

Commit 368d179adbac ("drm/i915/guc: Add GuC <-> kernel time stamp
translation information") added intel_device_info_print_runtime() in the
time info dump for no obvious reason or explanation in the commit
message. It only logs the rawclk freq. Remove it.

Cc: John Harrison <John.C.Harrison@Intel.com>
Cc: Alan Previn <alan.previn.teres.alexis@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/b395ac4c909042f5daabf29959d8733993545aa2.1660910433.git.jani.nikula@intel.com

Revert "drm/i915/guc: Add delay to disable scheduling after pin count goes to zero"

This reverts commit 6a079903847cce1dd06345127d2a32f26d2cd9c6.

Everything in CI using GuC is now timing out[1], and killing the machine
with this change (perhaps a deadlock?). CI was recently on fire due to
some changes coming in from -rc1, so likely the pre-merge CI results for
this series were invalid? For now just revert, unless GuC experts
already have a fix in mind.

[1] https://intel-gfx-ci.01.org/tree/drm-tip/index.html?

Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Alan Previn <alan.previn.teres.alexis@intel.com>
Cc: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220819123904.913750-1-matthew.auld@intel.com

drm/i915/guc: Add delay to disable scheduling after pin count goes to zero

Add a delay, configurable via debugfs (default 34ms), to disable
scheduling of a context after the pin count goes to zero. Disable
scheduling is a costly operation as it requires synchronizing with
the GuC. So the idea is that a delay allows the user to resubmit
something before doing this operation. This delay is only done if
the context isn't closed and less than a given threshold
(default is 3/4) of the guc_ids are in use.

As temporary WA disable this feature for the selftests. Selftests are
very timing sensitive and any change in timing can cause failure. A
follow up patch will fixup the selftests to understand this delay.

Alan Previn: Matt Brost first introduced this series back in Oct 2021.
However no real world workload with measured performance impact was
available to prove the intended results. Today, this series is being
republished in response to a real world workload that benefited greatly
from it along with measured performance improvement.

Workload description: 36 containers were created on a DG2 device where
each container was performing a combination of 720p 3d game rendering
and 30fps video encoding. The workload density was configured in a way
that guaranteed each container to ALWAYS be able to render and
encode no less than 30fps with a predefined maximum render + encode
latency time. That means the totality of all 36 containers and their
workloads were not saturating the engines to their max (in order to
maintain just enough headrooom to meet the min fps and max latencies
of incoming container submissions).

Problem statement: It was observed that the CPU core processing the i915
soft IRQ work was experiencing severe load. Using tracelogs and an
instrumentation patch to count specific i915 IRQ events, it was confirmed
that the majority of the CPU cycles were caused by the
gen11_other_irq_handler() -> guc_irq_handler() code path. The vast
majority of the cycles was determined to be processing a specific G2H
IRQ: i.e. INTEL_GUC_ACTION_SCHED_CONTEXT_MODE_DONE. These IRQs are sent
by GuC in response to i915 KMD sending H2G requests:
INTEL_GUC_ACTION_SCHED_CONTEXT_MODE_SET. Those H2G requests are sent
whenever a context goes idle so that we can unpin the context from GuC.
The high CPU utilization % symptom was limiting density scaling.

Root Cause Analysis: Because the incoming execution buffers were spread
across 36 different containers (each with multiple contexts) but the
system in totality was NOT saturated to the max, it was assumed that each
context was constantly idling between submissions. This was causing
a thrashing of unpinning contexts from GuC at one moment, followed quickly
by repinning them due to incoming workload the very next moment. These
event-pairs were being triggered across multiple contexts per container,
across all containers at the rate of > 30 times per sec per context.

Metrics: When running this workload without this patch, we measured an
average of ~69K INTEL_GUC_ACTION_SCHED_CONTEXT_MODE_DONE events every 10
seconds or ~10 million times over ~25+ mins. With this patch, the count
reduced to ~480 every 10 seconds or about ~28K over ~10 mins. The
improvement observed is ~99% for the average counts per 10 seconds.

Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220817020511.2180747-3-alan.previn.teres.alexis@intel.com

drm/i915/selftests: Use correct selfest calls for live tests

This will help in an upcoming patch where the live selftest wrappers
are extended to do more.

Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Reviewed-by: John Harrison <john.c.harrison@Intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220817020511.2180747-2-alan.previn.teres.alexis@intel.com

drm/i915/guc: clear stalled request after a reset

If the GuC CTs are full and we need to stall the request submission
while waiting for space, we save the stalled request and where the stall
occurred; when the CTs have space again we pick up the request submission
from where we left off.

If a full GT reset occurs, the state of all contexts is cleared and all
non-guilty requests are unsubmitted, therefore we need to restart the
stalled request submission from scratch. To make sure that we do so,
clear the saved request after a reset.

Fixes note: the patch that introduced the bug is in 5.15, but no
officially supported platform had GuC submission enabled by default
in that kernel, so the backport to that particular version (and only
that one) can potentially be skipped.

Fixes: 925dc1cf58ed ("drm/i915/guc: Implement GuC submission tasklet")
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: John Harrison <john.c.harrison@intel.com>
Cc: <stable@vger.kernel.org> # v5.15+
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220811210812.3239621-1-daniele.ceraolospurio@intel.com

drm/i915/guc: skip scrub_ctbs selftest if reset is disabled

The test needs GT reset to trigger the scrubbing logic, so we can only
run it when reset is enabled.

Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: John Harrison <john.c.harrison@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220708224158.929327-1-daniele.ceraolospurio@intel.com

drm/i915/guc: Reduce spam from error capture

Some debug code got left in when the GuC based register save for error
capture was added. Remove that.

Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220728022028.2190627-8-John.C.Harrison@Intel.com

drm/i915/guc: Make GuC log sizes runtime configurable

The GuC log buffer sizes had to be configured statically at compile
time. This can be quite troublesome when needing to get larger logs
out of a released driver. So re-organise the code to allow a boot time
module parameter override.

Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220728022028.2190627-7-John.C.Harrison@Intel.com

drm/i915/guc: Use streaming loads to speed up dumping the guc log

Use a temporary page and mempy_from_wc to reduce the time it takes to
dump the guc log to debugfs.

Signed-off-by: Chris Wilson <chris.p.wilson@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220728022028.2190627-6-John.C.Harrison@Intel.com

drm/i915/guc: Record CTB info in error logs

When debugging GuC communication issues, it is useful to have the CTB
info available. So add the state and buffer contents to the error
capture log.

Also, add a sub-structure for the GuC specific error capture info as
it is now becoming numerous.

Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220728022028.2190627-5-John.C.Harrison@Intel.com

drm/i915/guc: Add GuC <-> kernel time stamp translation information

It is useful to be able to match GuC events to kernel events when
looking at the GuC log. That requires being able to convert GuC
timestamps to kernel time. So, when dumping error captures and/or GuC
logs, include a stamp in both time zones plus the clock frequency.

Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220728022028.2190627-4-John.C.Harrison@Intel.com

drm/i915/guc: Fix capture size warning and bump the size

There was a size check to warn if the GuC error state capture buffer
allocation would be too small to fit a reasonable amount of capture
data for the current platform. Unfortunately, the test was done too
early in the boot sequence and was actually testing 'if(-ENODEV >
size)'.

Move the check to be later. The check is only used to print a warning
message, so it doesn't really matter how early or late it is done.
Note that it is not possible to dynamically size the buffer because
the allocation needs to be done before the engine information is
available (at least, it would be in the intended two-phase GuC init
process).

Now that the check works, it is reporting size too small for newer
platforms. The check includes a 3x oversample multiplier to allow for
multiple error captures to be bufferd by GuC before i915 has a chance
to read them out. This is less important than simply being big enough
to fit the first capture.

So a) bump the default size to be large enough for one capture minimum
and b) make the warning only if one capture won't fit, instead use a
notice for the 3x size.

Note that the size estimate is a worst case scenario. Actual captures
will likely be smaller.

Lastly, use drm_warn istead of DRM_WARN as the former provides more
infmration and the latter is deprecated.

Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220728022028.2190627-3-John.C.Harrison@Intel.com

drm/i915/guc: Add a helper for log buffer size

Add a helper to get GuC log buffer size.

Signed-off-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220728022028.2190627-2-John.C.Harrison@Intel.com

drm/i915/dg2: Add additional tuning settings

Some additional MMIO tuning settings have appeared in the bspec's
performance tuning guide section.

One of the tuning settings here is also documented as formal workaround
Wa_22012654132 for some steppings of DG2.  However the tuning setting
applies to all DG2 variants and steppings, making it a superset of the
workaround.

v2:
 - Move DRAW_WATERMARK to engine workaround section.  It only moves into
   the engine context on future platforms.  (Lucas)
 - CHICKEN_RASTER_2 needs to be handled as a masked register.  (Lucas)

Bspec: 68331
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220816210601.2041572-2-matthew.d.roper@intel.com

drm/i915/gt: Add dedicated function for non-ctx register tuning settings

The bspec performance tuning section gives recommended settings that the
driver should program for various MMIO registers.  Although these
settings aren't "workarounds" we use the workaround infrastructure to do
this programming to make sure it is handled at the appropriate places
and doesn't conflict with any real workarounds.

Since more of these are starting to show up on recent platforms, it's a
good time to create a dedicated function to hold them so that there's
less ambiguity about how/where to implement new ones.

Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220816210601.2041572-1-matthew.d.roper@intel.com

drm/i915/ttm: fix CCS handling

Crucible + recent Mesa seems to sometimes hit:

GEM_BUG_ON(num_ccs_blks > NUM_CCS_BLKS_PER_XFER)

And it looks like we can also trigger this with gem_lmem_swapping, if we
modify the test to use slightly larger object sizes.

Looking closer it looks like we have the following issues in
migrate_copy():

  - We are using plain integer in various places, which we can easily
    overflow with a large object.

  - We pass the entire object size (when the src is lmem) into
    emit_pte() and then try to copy it, which doesn't work, since we
    only have a few fixed sized windows in which to map the pages and
    perform the copy. With an object > 8M we therefore aren't properly
    copying the pages. And then with an object > 64M we trigger the
    GEM_BUG_ON(num_ccs_blks > NUM_CCS_BLKS_PER_XFER).

So it looks like our copy handling for any object > 8M (which is our
CHUNK_SZ) is currently broken on DG2.

Fixes: da0595ae91da ("drm/i915/migrate: Evict and restore the flatccs capable lmem obj")
Testcase: igt@gem_lmem_swapping
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Ramalingam C <ramalingam.c@intel.com>
Reviewed-by: Ramalingam C<ramalingam.c@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220805132240.442747-2-matthew.auld@intel.com

drm/i915/ttm: remove calc_ctrl_surf_instr_size

We only ever need to emit one ccs block copy command.

Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Ramalingam C <ramalingam.c@intel.com>
Reviewed-by: Ramalingam C<ramalingam.c@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220805132240.442747-1-matthew.auld@intel.com

drm/i915: pass a pointer for tlb seqno at vma_invalidate_tlb()

WRITE_ONCE() should happen at the original var, not on a local
copy of it.

Cc: stable@vger.kernel.org
Fixes: 5d36acb7198b ("drm/i915/gt: Batch TLB invalidations")
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
[added cc-stable while merging it]
Link: https://patchwork.freedesktop.org/patch/msgid/f9550e6bacea10131ff40dd8981b69eb9251cdcd.1659598090.git.mchehab@kernel.org

drm/i915/gem: Remove shared locking on freeing objects

The obj->base.resv may be shared across many objects, some of which may
still be live and locked, preventing objects from being freed
indefintely. We could individualise the lock during the free, or rely on
a freed object having no contention and being able to immediately free
the pages it owns.

References: https://gitlab.freedesktop.org/drm/intel/-/issues/6469
Fixes: be7612fd6665 ("drm/i915: Require object lock when freeing pages during destruction")
Fixes: 6cb12fbda1c2 ("drm/i915: Use trylock instead of blocking lock for __i915_gem_free_objects.")
Cc: <stable@vger.kernel.org> # v5.17+
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Tested-by: Nirmoy Das <nirmoy.das@intel.com>
Acked-by: Nirmoy Das <nirmoy.das@intel.com>
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220726144844.18429-1-nirmoy.das@intel.com

drm/i915/dg2: Add Wa_1509727124

Bspec: 46052
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Harish Chegondi <harish.chegondi@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220801213839.8549-1-harish.chegondi@intel.com

drm/i915/dg2: Update DG2 to GuC v70.4.1

New release of GuC with a bunch of fixes specific to DG2. Some of
these require follow up i915 changes to enable.

Note also that it is not necessary to maintain backwards compatibility
with 70.1.2 for DG2 because DG2 is still under force probe protection.

Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220728230722.2749701-2-John.C.Harrison@Intel.com

drm/i915/guc: Don't send policy update for child contexts.

The GuC FW applies the parent context policy to all the children,
so individual updates to the children are not supported and we
should not send them.

Note that sending the message did not have any functional consequences,
because the GuC just drops it and logs an error; since we were trying
to set the child policy to match the parent anyway the message being
dropped was not a problem.

Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: John Harrison <john.c.harrison@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220728003339.2361010-1-daniele.ceraolospurio@intel.com

drm/i915/guc: Don't abort on CTB_UNUSED status

When the KMD sends a CLIENT_RESET request to GuC (as part of the
suspend sequence), GuC will mark the CTB buffer as 'UNUSED'. If the
KMD then checked the CTB queue, it would see a non-zero status value
and report the buffer as corrupted.

Technically, no G2H messages should be received once the CLIENT_RESET
has been sent. However, if a context was outstanding on an engine then
it would get reset and a reset notification would be sent. So, don't
actually treat UNUSED as a catastrophic error. Just flag it up as
unexpected and keep going.

Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220728024225.2363663-7-John.C.Harrison@Intel.com

drm/i915/guc: Support larger contexts on newer hardware

The GuC needs a copy of a golden context for implementing watchdog
resets (aka media resets). This context is larger on newer platforms.
So adjust the size being allocated/copied accordingly.

Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220728024225.2363663-6-John.C.Harrison@Intel.com

drm/i915/selftest: Cope with not having an RCS engine

It is no longer guaranteed that there will always be an RCS engine.
So, use the helper function for finding the first available engine that
can be used for general purpose selftets.

Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220728024225.2363663-5-John.C.Harrison@Intel.com

drm/i915/guc: Add selftest for a hung GuC

Add a test to check that the hangcheck will recover from a submission
hang in the GuC.

Signed-off-by: Rahul Kumar Singh <rahul.kumar.singh@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220728182616.2417491-1-John.C.Harrison@Intel.com

drm/i915/guc: Fix issues with live_preempt_cancel

Having semaphores results in different behavior when a dependent request
is cancelled. In the case of semaphores the request could be on the HW
and complete successfully while without the request is held in the
driver and the error from the dependent request is propagated. Fix
live_preempt_cancel to take this behavior into account.

Also update live_preempt_cancel to use new function intel_context_ban
rather than intel_context_set_banned.

Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220728024225.2363663-3-John.C.Harrison@Intel.com

drm/i915/guc: Route semaphores to GuC for Gen12+

In GuC submission mode, there is an option to use auto-switch out
semaphores and have GuC auto-switch in a waiting context. This
requires routing the semaphore interrupt to GuC.

Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220728024225.2363663-2-John.C.Harrison@Intel.com

drm/i915/guc: Check for ct enabled while waiting for response

We are seeing error message of "No response for request". Some cases
happened while waiting for response and reset/suspend action was triggered.
In this case, no response is not an error, active requests will be
cancelled.

This patch will handle this condition and change the error message into
debug message.

Signed-off-by: Zhanjun Dong <zhanjun.dong@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/20220715211313.143645-1-zhanjun.dong@intel.com

drm/i915/gt: describe the new tlb parameter at i915_vma_resource

TLB cache invalidation can happen on two different situations:

1. synchronously, at __vma_put_pages();
2. asynchronously.

On the first case, TLB cache invalidation happens inside
__vma_put_pages(). So, no need to do it later on.

However, on the second case, the pages will keep in memory
until __i915_vma_evict() is called.

So, we need to store the TLB data at struct i915_vma_resource,
in order to do a TLB cache invalidation before allowing
userspace to re-use the same memory.

So, i915_vma_resource_unbind() has gained a new parameter
in order to store the TLB data at the second case.

Document it.

Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/aa55eef7e63b8f3d0f69b525db2dd2eb87e9db6b.1658924372.git.mchehab@kernel.org

drm/i915/gt: Batch TLB invalidations

Invalidate TLB in batches, in order to reduce performance regressions.

Currently, every caller performs a full barrier around a TLB
invalidation, ignoring all other invalidations that may have already
removed their PTEs from the cache. As this is a synchronous operation
and can be quite slow, we cause multiple threads to contend on the TLB
invalidate mutex blocking userspace.

We only need to invalidate the TLB once after replacing our PTE to
ensure that there is no possible continued access to the physical
address before releasing our pages. By tracking a seqno for each full
TLB invalidate we can quickly determine if one has been performed since
rewriting the PTE, and only if necessary trigger one for ourselves.

That helps to reduce the performance regression introduced by TLB
invalidate logic.

[mchehab: rebased to not require moving the code to a separate file]

Cc: stable@vger.kernel.org
Fixes: 7938d61591d3 ("drm/i915: Flush TLBs before releasing backing store")
Suggested-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Chris Wilson <chris.p.wilson@intel.com>
Cc: Fei Yang <fei.yang@intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/4e97ef5deb6739cadaaf40aa45620547e9c4ec06.1658924372.git.mchehab@kernel.org

drm/i915/gt: Skip TLB invalidations once wedged

Skip all further TLB invalidations once the device is wedged and
had been reset, as, on such cases, it can no longer process instructions
on the GPU and the user no longer has access to the TLB's in each engine.

So, an attempt to do a TLB cache invalidation will produce a timeout.

That helps to reduce the performance regression introduced by TLB
invalidate logic.

Cc: stable@vger.kernel.org
Fixes: 7938d61591d3 ("drm/i915: Flush TLBs before releasing backing store")
Signed-off-by: Chris Wilson <chris.p.wilson@intel.com>
Cc: Fei Yang <fei.yang@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Acked-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/5aa86564b9ec5fe7fe605c1dd7de76855401ed73.1658924372.git.mchehab@kernel.org

drm/i915/gt: Invalidate TLB of the OA unit at TLB invalidations

Ensure that the TLB of the OA unit is also invalidated
on gen12 HW, as just invalidating the TLB of an engine is not
enough.

Cc: stable@vger.kernel.org
Fixes: 7938d61591d3 ("drm/i915: Flush TLBs before releasing backing store")
Signed-off-by: Chris Wilson <chris.p.wilson@intel.com>
Cc: Fei Yang <fei.yang@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Acked-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/59724d9f5cf1e93b1620d01b8332ac991555283d.1658924372.git.mchehab@kernel.org

drm/i915/gt: document with_intel_gt_pm_if_awake()

Add a kernel-doc markup to document this new macro.

Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/b974905bd0f6b5308b91561cc85eeecd94f1452a.1658924372.git.mchehab@kernel.org

drm/i915/gt: Ignore TLB invalidations on idle engines

Check if the device is powered down prior to any engine activity,
as, on such cases, all the TLBs were already invalidated, so an
explicit TLB invalidation is not needed, thus reducing the
performance regression impact due to it.

This becomes more significant with GuC, as it can only do so when
the connection to the GuC is awake.

Cc: stable@vger.kernel.org
Fixes: 7938d61591d3 ("drm/i915: Flush TLBs before releasing backing store")
Signed-off-by: Chris Wilson <chris.p.wilson@intel.com>
Cc: Fei Yang <fei.yang@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Acked-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/278a57a672edac75683f0818b292e95da583a5fe.1658924372.git.mchehab@kernel.org

drm/i915/ttm: don't leak the ccs state

The kernel only manages the ccs state with lmem-only objects, however
the kernel should still take care not to leak the CCS state from the
previous user.

Fixes: 48760ffe923a ("drm/i915/gt: Clear compress metadata for Flat-ccs objects")
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Ramalingam C <ramalingam.c@intel.com>
Reviewed-by: Ramalingam C <ramalingam.c@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220727164346.282407-1-matthew.auld@intel.com

drm/i915: disable pci resize on 32-bit machine

PCI bar resize only works with 64 bit BAR so disable
this on 32-bit machine and resolve below compilation error:

drivers/gpu/drm/i915/gt/intel_region_lmem.c:94:23: error: result of
comparison of constant 4294967296 with expression of type
'resource_size_t' (aka 'unsigned int') is always false
[-Werror,-Wtautological-constant-out-of-range-compare]
                    root_res->start > 0x100000000ull)

Fixes: a91d1a17cd341 ("drm/i915: Add support for LMEM PCIe resizable bar")
Reported-by: Linux Kernel Functional Testing <lkft@linaro.org>
Tested-by: Linux Kernel Functional Testing <lkft@linaro.org>
Acked-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220727173306.16247-1-nirmoy.das@intel.com

drm/i915: Suppress oom warning for shmemfs object allocation failure

We report object allocation failures to userspace with ENOMEM, yet we
still show the memory warning after failing to shrink device allocated
pages. While this warning is similar to other system page allocation
failures, it is superfluous to the ENOMEM provided directly to
userspace.

v2: Add NOWARN in few more places from where we might return
    ENOMEM to userspace.

Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/4936
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Co-developed-by: Nirmoy Das <nirmoy.das@intel.com>
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220727174023.16766-1-nirmoy.das@intel.com

drm/i915/selftests: Fix comment typo

Fix the double `wait' typo in comment.

Signed-off-by: Jason Wang <wangborong@cdjrlc.com>
Reviewed-by: Andrzej Hajda <andrzej.hajda@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220716040520.31676-1-wangborong@cdjrlc.com

drm/i915/gt: Remove unneeded semicolon

The semicolon after the `}' is unneeded.

Signed-off-by: Jason Wang <wangborong@cdjrlc.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
[Removed line mention when pushing]
Link: https://patchwork.freedesktop.org/patch/msgid/20220716184439.72056-1-wangborong@cdjrlc.com

drm/i915/guc: Don't use pr_err when not necessary

A bunch of code was copy/pasted using pr_err as the default way to
report errors. However, drm_err is significantly more useful in
identifying where the error came from. So update the code to use that
instead.

Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220715004028.2126239-1-John.C.Harrison@Intel.com

drm/i915/guc: support v69 in parallel to v70

This patch re-introduces support for GuC v69 in parallel to v70. As this
is a quick fix, v69 has been re-introduced as the single "fallback" guc
version in case v70 is not available on disk and only for platforms that
are out of force_probe and require the GuC by default. All v69 specific
code has been labeled as such for easy identification, and the same was
done for all v70 functions for which there is a separate v69 version,
to avoid accidentally calling the wrong version via the unlabeled name.

When the fallback mode kicks in, a drm_notice message is printed in
dmesg to inform the user of the required update. The existing
logging of the fetch function has also been updated so that we no
longer complain immediately if we can't find a fw and we only throw an
error if the fetch of both the base and fallback blobs fails.

The plan is to follow this up with a more complex rework to allow for
multiple different GuC versions to be supported at the same time.

v2: reduce the fallback to platform that require it, switch to
firmware_request_nowarn(), improve logs.

Fixes: 2584b3549f4c ("drm/i915/guc: Update to GuC version 70.1.1")
Link: https://lists.freedesktop.org/archives/intel-gfx/2022-July/301640.html
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: John Harrison <John.C.Harrison@Intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Cc: Dave Airlie <airlied@gmail.com>
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Acked-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220718230732.1409641-1-daniele.ceraolospurio@intel.com

drm/i915/gt: Expose per-gt RPS defaults in sysfs

Add the following sysfs files to gt/gtN/.defaults/:
* rps_min_freq_mhz
* rps_max_freq_mhz

v2: Correct gt/gtN/.defaults/* file names in commit message
v3: Remove rps_boost_freq_mhz since it is not consumed by userspace

Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Cc: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/cf6e483bf79f871c2c8c74af6005bf6a83a3a1ce.1658192398.git.ashutosh.dixit@intel.com

drm/i915/gt: Create gt/gtN/.defaults/ for per gt sysfs defaults

Create a gt/gtN/.defaults/ directory (similar to
engine/<engine-name>/.defaults/) to expose default parameter values for
each gt in sysfs. This allows userspace to restore default parameter values
after they have changed. The empty 'struct gt_defaults' will be populated
by subsequent patches.

v2: Changed 'struct intel_rps_defaults rps_defaults' to
    'struct gt_defaults defaults' (Andi)

Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Cc: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/be7c30d0ae58be9d8d5b8242ba00a1b2825e63ad.1658192398.git.ashutosh.dixit@intel.com

drm/i915/reset: Handle reset timeouts under unrelated kernel hangs

When resuming after hibernate sometimes we see hangs in unrelated kernel
subsystems. These hangs often result in the following i915 trace:

i915 0000:00:02.0: [drm] *ERROR* \
intel_gt_reset_global timed out, cancelling all in-flight rendering

implying our reset task has been starved by the hanging kernel subsystem,
causing us to inappropiately declare the system as wedged beyond recovery.

The trace would be caused by our synchronize_srcu_expedited() taking more
than the allowed 5s due to the unrelated kernel hang. But we neither need
to perform that synchronisation inside the reset watchdog, nor do we need
such a short timeout before declaring the device as unrecoverable.

v2: Restore watchdog timeout to the previous 5 seconds (Ashutosh)

Bug: https://gitlab.freedesktop.org/drm/intel/-/issues/3575
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220630043959.5708-1-ashutosh.dixit@intel.com

drm/i915: Add lmem_bar_size modparam

For testing purposes, support forcing the lmem_bar_size through a new
modparam. In CI we only have a limited number of configurations for DG2,
but we still need to be reasonably sure we get a usable device (also
verifying we report the correct values for things like
probed_cpu_visible_size etc) with all the potential lmem_bar sizes that
we might expect see in the wild.

v2: Update commit message and a minor modification.(Matt)

v3: Optimised lmem bar size code and modified code to resize
bar maximum upto lmem_size instead of maximum supported size.(Nirmoy)

v4: Optimised lmem bar size code.(Nirmoy)

Signed-off-by: Priyanka Dandamudi <priyanka.dandamudi@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220713130209.2573233-3-priyanka.dandamudi@intel.com

drm/i915: Add support for LMEM PCIe resizable bar

Add support for the local memory PICe resizable bar, so that
local memory can be resized to the maximum size supported by the device,
and mapped correctly to the PCIe memory bar. It is usual that GPU
devices expose only 256MB BARs primarily to be compatible with 32-bit
systems. So, those devices cannot claim larger memory BAR windows size due
to the system BIOS limitation. With this change, it would be possible to
reprogram the windows of the bridge directly above the requesting device
on the same BAR type.

v2:Moved code to gt/intel_region_lmem.c and used only
single underscore for function names.(Jani)

v3: Optimised code.

Signed-off-by: Akeem G Abodunrin <akeem.g.abodunrin@intel.com>
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
Cc: Stuart Summers <stuart.summers@intel.com>
Cc: Michael J Ruhl <michael.j.ruhl@intel.com>
Cc: Prathap Kumar Valsan <prathap.kumar.valsan@intel.com>
Cc: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Priyanka Dandamudi <priyanka.dandamudi@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220713130209.2573233-2-priyanka.dandamudi@intel.com

drm/i915: Correct ss -> steering calculation for pre-Xe_HP platforms

Accidental use of a "SLICE" macro where a "SUBSLICE" macro was intended
causes the group ID for steering to be calculated incorrectly on
pre-Xe_HP platforms.

Fixes: 9a92732f040a ("drm/i915/gt: Add general DSS steering iterator to intel_gt_mcr")
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220712220513.3451794-1-matthew.d.roper@intel.com

drm/i915/ttm: fix 32b build

Since segment_pages is no longer a compile time constant, it looks the
DIV_ROUND_UP(node->size, segment_pages) breaks the 32b build. Simplest
is just to use the ULL variant, but really we should need not need more
than u32 for the page alignment (also we are limited by that due to the
sg->length type), so also make it all u32.

Reported-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Fixes: bc99f1209f19 ("drm/i915/ttm: fix sg_table construction")
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Nirmoy Das <nirmoy.das@linux.intel.com>
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220712174050.592550-1-matthew.auld@intel.com

drm/i915/selftests: fix subtraction overflow bug

On some machines hole_end can be small enough to cause subtraction
overflow. On the other side (addr + 2 * min_alignment) can overflow
in case of mock tests. This patch should handle both cases.

Fixes: e1c5f754067b59 ("drm/i915: Avoid overflow in computing pot_hole loop termination")
Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/3674
Signed-off-by: Andrzej Hajda <andrzej.hajda@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220624113528.2159210-1-andrzej.hajda@intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/i915/gt: Only kick the signal worker if there's been an update

One impact of commit 047a1b877ed4 ("dma-buf & drm/amdgpu: remove
dma_resv workaround") is that it stores many, many more fences. Whereas
adding an exclusive fence used to remove the shared fence list, that
list is now preserved and the write fences included into the list. Not
just a single write fence, but now a write/read fence per context. That
causes us to have to track more fences than before (albeit half of those
are redundant), and we trigger more interrupts for multi-engine
workloads.

As part of reducing the impact from handling more signaling, we observe
we only need to kick the signal worker after adding a fence iff we have
good cause to believe that there is work to be done in processing the
fence i.e. we either need to enable the interrupt or the request is
already complete but we don't know if we saw the interrupt and so need
to check signaling.

References: 047a1b877ed4 ("dma-buf & drm/amdgpu: remove dma_resv workaround")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Karolina Drobnik <karolina.drobnik@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/d7b953c7a4ba747c8196a164e2f8c5aef468d048.1657289332.git.karolina.drobnik@intel.com

drm/i915: Bump GT idling delay to 2 jiffies

In monitoring a transcode pipeline that is latency sensitive (it waits
between submitting frames, and each frame requires work on rcs/vcs/vecs
engines), it is found that it took longer than a single jiffy for it to
sustain its workload. Allowing an extra jiffy headroom for the userspace
prevents us from prematurely parking and having to exit powersaving
immediately.

Link: https://gitlab.freedesktop.org/drm/intel/-/issues/6284
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Karolina Drobnik <karolina.drobnik@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/e37911ec087a9ce50630d6faf61fa2c0d5f96d44.1657289332.git.karolina.drobnik@intel.com

drm/i915/gem: Look for waitboosting across the whole object prior to individual waits

We employ a "waitboost" heuristic to detect when userspace is stalled
waiting for results from earlier execution. Under latency sensitive work
mixed between the gpu/cpu, the GPU is typically under-utilised and so
RPS sees that low utilisation as a reason to downclock the frequency,
causing longer stalls and lower throughput. The user left waiting for
the results is not impressed.

On applying commit 047a1b877ed4 ("dma-buf & drm/amdgpu: remove dma_resv
workaround") it was observed that deinterlacing h264 on Haswell
performance dropped by 2-5x. The reason being that the natural workload
was not intense enough to trigger RPS (using HW evaluation intervals) to
upclock, and so it was depending on waitboosting for the throughput.

Commit 047a1b877ed4 ("dma-buf & drm/amdgpu: remove dma_resv workaround")
changes the composition of dma-resv from keeping a single write fence +
multiple read fences, to a single array of multiple write and read
fences (a maximum of one pair of write/read fences per context). The
iteration order was also changed implicitly from all-read fences then
the single write fence, to a mix of write fences followed by read
fences. It is that ordering change that belied the fragility of
waitboosting.

Currently, a waitboost is inspected at the point of waiting on an
outstanding fence. If the GPU is backlogged such that we haven't yet
stated the request we need to wait on, we force the GPU to upclock until
the completion of that request. By changing the order in which we waited
upon requests, we ended up waiting on those requests in sequence and as
such we saw that each request was already started and so not a suitable
candidate for waitboosting.

Instead of asking whether to boost each fence in turn, we can look at
whether boosting is required for the dma-resv ensemble prior to waiting
on any fence, making the heuristic more robust to the order in which
fences are stored in the dma-resv.

Reported-by: Thomas Voegtle <tv@lio96.de>
Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/6284
Fixes: 047a1b877ed4 ("dma-buf & drm/amdgpu: remove dma_resv workaround")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Karolina Drobnik <karolina.drobnik@intel.com>
Tested-by: Thomas Voegtle <tv@lio96.de>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Acked-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/07e05518d9f6620d20cc1101ec1849203fe973f9.1657289332.git.karolina.drobnik@intel.com

drm/i915/gt: Serialize TLB invalidates with GT resets

Avoid trying to invalidate the TLB in the middle of performing an
engine reset, as this may result in the reset timing out. Currently,
the TLB invalidate is only serialised by its own mutex, forgoing the
uncore lock, but we can take the uncore->lock as well to serialise
the mmio access, thereby serialising with the GDRST.

Tested on a NUC5i7RYB, BIOS RYBDWi35.86A.0380.2019.0517.1530 with
i915 selftest/hangcheck.

Cc: stable@vger.kernel.org # v4.4 and upper
Fixes: 7938d61591d3 ("drm/i915: Flush TLBs before releasing backing store")
Reported-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Tested-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Reviewed-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Signed-off-by: Chris Wilson <chris.p.wilson@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Acked-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/1e59a7c45dd919a530256b9ac721ac6ea86c0677.1657639152.git.mchehab@kernel.org

drm/i915/gt: Serialize GRDOM access between multiple engine resets

Don't allow two engines to be reset in parallel, as they would both
try to select a reset bit (and send requests to common registers)
and wait on that register, at the same time. Serialize control of
the reset requests/acks using the uncore->lock, which will also ensure
that no other GT state changes at the same time as the actual reset.

Cc: stable@vger.kernel.org # v4.4 and upper
Reported-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Andi Shyti <andi.shyti@intel.com>
Reviewed-by: Andrzej Hajda <andrzej.hajda@intel.com>
Acked-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/e0a2d894e77aed7c2e36b0d1abdc7dbac3011729.1657639152.git.mchehab@kernel.org

drm/i915/dg2: Add Wa_15010599737

This workaround may need to be extended to other platforms soon, but for
now it's marked as DG2-specific.

Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Arun R Murthy <arun.r.murthy@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220708215804.2889246-1-matthew.d.roper@intel.com

drm/i915/ttm: fix sg_table construction

If we encounter some monster sized local-memory page that exceeds the
maximum sg length (UINT32_MAX), ensure that don't end up with some
misaligned address in the entry that follows, leading to fireworks
later. Also ensure we have some coverage of this in the selftests.

v2(Chris):
  - Use round_down consistently to avoid udiv errors
v3(Nirmoy):
  - Also update the max_segment in the selftest

Fixes: f701b16d4cc5 ("drm/i915/ttm: add i915_sg_from_buddy_resource")
Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/6379
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Nirmoy Das <nirmoy.das@linux.intel.com>
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220711085859.24198-1-matthew.auld@intel.com

drm/i915/selftests: fix a couple IS_ERR() vs NULL tests

The shmem_pin_map() function doesn't return error pointers, it returns
NULL.

Fixes: be1cb55a07bf ("drm/i915/gt: Keep a no-frills swappable copy of the default context state")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220708094104.GL2316@kadam

drm/i915/mtl: Add MeteorLake PCI IDs

Add Meteorlake PCI IDs. Split into M, and P subplatforms.

v2: Update PCI id's
v3: Move id 7d60 under MTL_M(MattR)

Bspec: 55420

Signed-off-by: Radhakrishna Sripada <radhakrishna.sripada@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@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/20220708000335.2869311-3-radhakrishna.sripada@intel.com

drm/i915/mtl: Add MeteorLake platform info

MTL has Xe_LPD+ display IP (version = 14), MTL graphics IP
(version = 12.70), and Xe_LPM+ media IP (version = 13).

Bspec: 55413
Bspec: 55416
Bspec: 55417
Bspec: 55418
Bspec: 55726
Bspec: 45544
Bspec: 65380

v2: rearrange the fields in pci_info(MattR)

Cc: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Radhakrishna Sripada <radhakrishna.sripada@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
[mattrope: Moved IS_METEORLAKE() higher in header]
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220708000335.2869311-2-radhakrishna.sripada@intel.com

drm/i915/gt: Add general DSS steering iterator to intel_gt_mcr

Although all DSS belong to a single pool on Xe_HP platforms (i.e.,
they're not organized into slices from a topology point of view), we do
still need to pass 'group' and 'instance' targets when steering register
accesses to a specific instance of a per-DSS multicast register.  The
rules for how to determine group and instance IDs (which previously used
legacy terms "slice" and "subslice") varies by platform.  Some platforms
determine steering by gslice membership, some platforms by cslice
membership, and future platforms may have other rules.

Since looping over each DSS and performing steered unicast register
accesses is a relatively common pattern, let's add a dedicated iteration
macro to handle this (and replace the platform-specific "instdone" loop
we were using previously.  This will avoid the calling code needing to
figure out the details about how to obtain steering IDs for a specific
DSS.

Most of the places where we use this new loop are in the GPU errorstate
code at the moment, but we do have some additional features coming in
the future that will also need to loop over each DSS and steer some
register accesses accordingly.

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/20220701232006.1016135-1-matthew.d.roper@intel.com

i915/perf: Disable OA sseu config param for gfx12.50+

The global sseu config is applicable only to gen11 platforms where
concurrent media, render and OA use cases may cause some subslices to be
turned off and hence lose NOA configuration. Ideally we want to return
ENODEV for non-gen11 platforms, however, this has shipped with gfx12, so
disable only for gfx12.50+.

v2: gfx12 is already shipped with this, disable for gfx12.50+ (Lionel)

v3: (Matt)
- Update commit message and replace "12.5" with "12.50"
- Replace DRM_DEBUG() with driver specific drm_dbg()

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Andrzej Hajda <andrzej.hajda@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220707193002.2859653-2-umesh.nerlige.ramappa@intel.com

i915/perf: Replace DRM_DEBUG with driver specific drm_dbg call

DRM_DEBUG is not the right debug call to use in i915 OA, replace it with
driver specific drm_dbg() call (Matt).

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Andrzej Hajda <andrzej.hajda@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220707193002.2859653-1-umesh.nerlige.ramappa@intel.com

drm/i915/selftests: Grab the runtime pm in shrink_thp

Since we are not holding a wakeref, shrinking a bound object is not
guaranteed.

Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/6370
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220706154738.235204-1-matthew.auld@intel.com

drm/i915/guc: Asynchronous flush of GuC log regions

Both error-capture and relay-logging mechanism use the GuC
log infrastructure. That means the KMD must send a log flush
complete notification back to GuC after reading the data out.
This call is currently being sent synchronously.
However, synchronous H2Gs cause problems when the system is
backed up. There is no need for this to be synchronous. The
KMD wasn't even looking at the return status from it. So make
it asynchronous and then there is no issue about time outs.

Signed-off-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220607002314.1451656-2-alan.previn.teres.alexis@intel.com

drm/i915: Fix vm use-after-free in vma destruction

In vma destruction, the following race may occur:

Thread 1:        Thread 2:
i915_vma_destroy();

  ...
  list_del_init(vma->vm_link);
  ...
  mutex_unlock(vma->vm->mutex);
  __i915_vm_release();
release_references();

And in release_reference() we dereference vma->vm to get to the
vm gt pointer, leading to a use-after free.

However, __i915_vm_release() grabs the vm->mutex so the vm won't be
destroyed before vma->vm->mutex is released, so extract the gt pointer
under the vm->mutex to avoid the vma->vm dereference in
release_references().

v2: Fix a typo in the commit message (Andi Shyti)

Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/5944
Fixes: e1a7ab4fca0c ("drm/i915: Remove the vm open count")

Cc: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Acked-by: Nirmoy Das <nirmoy.das@intel.con>
Reviewed-by: Andrzej Hajda <andrzej.hajda@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220620123659.381772-1-thomas.hellstrom@linux.intel.com

drm/doc/rfc: VM_BIND uapi definition

VM_BIND and related uapi definitions

v2: Reduce the scope to simple Mesa use case.
v3: Expand VM_UNBIND documentation and add
    I915_GEM_VM_BIND/UNBIND_FENCE_VALID
    and I915_GEM_VM_BIND_TLB_FLUSH flags.
v4: Remove I915_GEM_VM_BIND_TLB_FLUSH flag and add additional
    documentation for vm_bind/unbind.
v5: Remove TLB flush requirement on VM_UNBIND.
    Add version support to stage implementation.
v6: Define and use drm_i915_gem_timeline_fence structure for
    all timeline fences.
v7: Rename I915_PARAM_HAS_VM_BIND to I915_PARAM_VM_BIND_VERSION.
    Update documentation on async vm_bind/unbind and versioning.
    Remove redundant vm_bind/unbind FENCE_VALID flag, execbuf3
    batch_count field and I915_EXEC3_SECURE flag.
v8: Remove I915_GEM_VM_BIND_READONLY and minor documentation
    updates.

Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Acked-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220701003110.24843-4-niranjana.vishwanathapura@intel.com

drm/i915: Update i915 uapi documentation

Add some missing i915 uapi documentation which the new
i915 VM_BIND feature documentation will be refer to.

Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220701003110.24843-3-niranjana.vishwanathapura@intel.com

drm/doc/rfc: VM_BIND feature design document

VM_BIND design document with description of intended use cases.

v2: Reduce the scope to simple Mesa use case.
v3: Expand documentation on dma-resv usage, TLB flushing and
    execbuf3.
v4: Remove vm_bind tlb flush request support.
v5: Update TLB flushing documentation.
v6: Update out of order completion documentation.

Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Acked-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220701003110.24843-2-niranjana.vishwanathapura@intel.com

drm/i915: DG2 and ATS-M device ID updates

Small BAR support has now landed, which allows us to add the PCI IDs
that correspond to add-in card designs of DG2 and ATS-M.  There's also
one additional MB-down PCI ID that recently appeared (0x5698) so we add
it too.

Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220701152231.529511-2-matthew.d.roper@intel.com

drm/i915/pvc: Implement w/a 16016694945

A new PVC-specific workaround has just been added to the BSpec.

BSpec: 64027

Signed-off-by: Gustavo Sousa <gustavo.sousa@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/20220630201407.16770-1-gustavo.sousa@intel.com