On i965gm the hardware frame counter does not work when
the TV encoder is active. So let's not try to consult
the hardware frame counter in that case. Instead we'll
fall back to the timestamp based guesstimation method
used on gen2.
Note that the pipe timings generated by the TV encoder
are also rather peculiar. Apparently the pipe wants to
run at a much higher speed (related to the oversample
clock somehow it seems) but during the vertical active
period the TV encoder stalls the pipe every few lines
to keep its speed in check. But once the vertical
blanking period is reached the pipe gets to run at full
speed. This means our vblank timestamp estimates are
suspect. Fixing all that would require quite a bit
more work. This simple fix at least avoids the nasty
vblank timeouts that are happening currently.
Curiously the frame counter works just fine on i945gm
and gm45. I don't really understand what kind of mishap
occurred with the hardware design on i965gm. Sadly
I wasn't able to find any chicken bits etc. that would
fix the frame counter :(
v2: Move the zero vs. non-zero hw counter value handling
into i915_get_vblank_counter() (Daniel)
Use the per-crtc maximum exclusively, leaving the
per-device maximum at zero
v3: max_vblank_count not populated yet in intel_enable_pipe()
use intel_crtc_max_vblank_count() instead
Cc: stable@vger.kernel.org
Cc: Daniel Vetter <daniel@ffwll.ch>
Fixes: 51e31d49c8 ("drm/i915: Use generic vblank wait")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93782
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190122125149.GE5527@ideak-desk.fi.intel.com
Reviewed-by: Imre Deak <imre.deak@intel.com>
Now that the submission backends are controlled via their own spinlocks,
with a wave of a magic wand we can lift the struct_mutex requirement
around GPU reset. That is we allow the submission frontend (userspace)
to keep on submitting while we process the GPU reset as we can suspend
the backend independently.
The major change is around the backoff/handoff strategy for performing
the reset. With no mutex deadlock, we no longer have to coordinate with
any waiter, and just perform the reset immediately.
Testcase: igt/gem_mmap_gtt/hang # regresses
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190125132230.22221-3-chris@chris-wilson.co.uk
The guc (and huc) currently inexcruitably depend on struct_mutex for
device reinitialisation from inside the reset, and indeed taking any
mutex here is verboten (as we must be able to reset from underneath any
of our mutexes). That makes recovering the guc unviable without, for
example, reserving contiguous vma space and pages for it to use.
The plan to re-enable global reset for the GuC centres around reusing the
WOPM reserved space at the top of the aperture (that we know we can
populate a contiguous range large enough to dma xfer the fw image).
In the meantime, hopefully no one even notices as the device-reset is
only used as a backup to the per-engine resets for handling GPU hangs.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Acked-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190125132230.22221-2-chris@chris-wilson.co.uk
Simplify by using sizeof(u32) to convert from the index inside the HWSP
to the byte offset. This has the advantage of not only being shorter
(and so not upsetting checkpatch!) but that it matches use where we are
writing to byte addresses using other commands than MI_STORE_DWORD_IMM.
v2: Drop the now superfluous MI_STORE_DWORD_INDEX_SHIFT, it appears to
be a local invention so keeping it after the final use does not help to
clarify the GPU instruction.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190125120005.25191-2-chris@chris-wilson.co.uk
The table has been unified across OSes to minimize virtualization overhead.
The MOCS table is now published as part of bspec, and versioned. Entries
are supposed to never be modified, but new ones can be added. Adding
entries increases table version. The patch includes version 1 entries.
Meaning of each entry is now explained in bspec, and user mode clients
are expected to know what each entry means. The 3 entries used for previous
platforms are still compatible with their legacy definitions, but that is
not guaranteed to be true for future platforms.
v2: Fixed SCC values, improved commit comment (Daniele)
v3: Improved MOCS table comment (Daniele)
v4: Moved new entries below gen9 ones. Put common entries into
definition to be used in multiple arrays. (Lucas)
v5: Made defines for or-ing flags. Renamed macros from MOCS_TABLE
to MOCS_ENTRIES. Switched LE_CoS to upper case. (Joonas)
v6: Removed definitions of reserved entries. (Michal)
Increased limit of entries sent to the hardware on gen11+.
v7: Simplify table as done for previou gens (Lucas)
v8: Rebase on cached number of entries per-platform and use new
MOCS_ENTRY() macro (Lucas)
v9: Update comment (from Tomasz)
BSpec: 34007
BSpec: 560
Signed-off-by: Tomasz Lis <tomasz.lis@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190124000604.18861-8-lucas.demarchi@intel.com
Instead of considering we have defined entries for any index in the
table, let's keep track of the ones we explicitly defined. This will
allow Gen 11 to have it's new table defined in which we have holes of
undefined entries.
Repeated comments about the meaning of undefined entries were removed
since they are overly verbose and copy-pasted in several functions: now
the definition is in the top only.
v2: add helper function to get the index (from Chris)
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190124000604.18861-6-lucas.demarchi@intel.com
The MOCS tables are going to be very similar across platforms.
To reduce the amount of copied code, this patch rips the common part and
puts it into a definition valid for all gen9 platforms.
v2: Made defines for or-ing flags. Renamed macros from MOCS_TABLE
to MOCS_ENTRIES. (Joonas)
v3 (Lucas):
- Fix indentation
- Rebase on rework done by additional patch
- Remove define for or-ing flags as it made the table more complex by
requiring zeroed values to be passed
- Do not embed comma in the macro, so to treat that just as another
item and please source code formatting tools
Signed-off-by: Tomasz Lis <tomasz.lis@intel.com>
Suggested-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190124000604.18861-4-lucas.demarchi@intel.com
Make the defines for LE and L3 caching options to contain the shifts and
remove the zeros from the tables as shifting zeros always result in
zero.
Starting from Ice Lake the MOCS table is defined in the spec and
contains all entries. So to simplify checking the table with the values
set in code, the value is now part of the macro name. This allows to
still give the most used option and sensible name, but also to easily
cross check the table from the spec for gen >= 11.
By removing the zeros we avoid maintaining a huge table since the one
from spec contains many more entries. The new table for Ice Lake will
be added by other patches, this only reformats the table.
While at it also fix the indentation.
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Tomasz Lis <tomasz.lis@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190124000604.18861-3-lucas.demarchi@intel.com
Instead of initializing them to uncached, let's set them to PTE for
kernel tracking. While at it do some minor adjustments to comments and
coding style.
From Chris: "What it does mean is that the buffer contents are consistent
with our cache tracking; and for userspace the results were always
undefined. So we should at least be able to guarantee that the data
written by userspace from the CPU is visible. After that, your caches
are on your own".
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tomasz Lis <tomasz.lis@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190124000604.18861-2-lucas.demarchi@intel.com
Now that our state comparison functions are pretty complete, we should
enable fastset by default when a modeset can be avoided. Even if we're
not completely certain about the inherited state, we can be certain
after the first modeset that our sw state matches the hw state.
There is one testcase explicitly testing fastset,
kms_panel_fitting.atomic-fastset but other testcases do so indirectly
because most tests don't clean up the display during exit, or otherwise
indirectly preserve mode by doing igt_display_reset or inheriting during
init.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
[mlankhorst: Use DRM_DEBUG_KMS. (j4ni)]
Link: https://patchwork.freedesktop.org/patch/msgid/20190108160842.13396-3-maarten.lankhorst@linux.intel.com
With new platforms not having LVDS support, only call intel_lvds_init()
on platforms that might actually have LVDS. Move the comment about eDP
init to the PCH block where it's relevant.
This puts intel_lvds_init() more in line with the rest of the outputs,
and makes it slightly easier for the uninitiated to figure out which
platforms actually have what.
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190122082307.4003-2-jani.nikula@intel.com
With new platforms not having CRT support and most conditions in
intel_crt_present() being specific to DDI, split out the CRT
initialization to platform specific blocks in the if ladder. Add new
Pineview block for this.
This puts intel_crt_init() more in line with the rest of the outputs,
and makes it slightly easier for the uninitiated to figure out which
platforms actually have what.
v2: keep gen >= 9 check in intel_ddi_crt_present() (Ville)
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190122082307.4003-1-jani.nikula@intel.com
We currently program userspace-provided gamma and degamma LUT's into our
hardware without really checking to see whether they satisfy our
hardware's rules. We should try to catch tables that are invalid for
our hardware early and reject the atomic transaction.
All of our platforms that accept a degamma LUT expect that the entries
in the LUT are always flat or increasing, never decreasing. Also, our
GLK and ICL platforms only accept degamma tables with r=g=b entries; so
we should also add the relevant checks for that in anticipation of
degamma support landing for those platforms.
v2:
- Use new API (single check function with bitmask of tests to apply)
- Call helper for our gamma table as well (with no additional tests
specified) so that the table size will be validated.
v3:
- Don't call on the gamma table since the LUT size is already tested at
property blob upload and we don't have any additional hardware
constraints for that LUT.
v4:
- Apply equal color channel check on gen10 as well; the bspec has some
strange tagging for CNL platforms, but this appears to apply there as
well. (Ville)
Cc: Uma Shankar <uma.shankar@intel.com>
Cc: Swati Sharma <swati2.sharma@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Uma Shankar <uma.shankar@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181218175158.5739-1-matthew.d.roper@intel.com
Some hardware may place additional restrictions on the gamma/degamma
curves described by our LUT properties. E.g., that a gamma curve never
decreases or that the red/green/blue channels of a LUT's entries must be
equal. Let's add a helper function that drivers can use to test that a
userspace-provided LUT is valid and doesn't violate hardware
requirements.
v2:
- Combine into a single helper that just takes a bitmask of the tests
to apply. (Brian Starkey)
- Add additional check (always performed) that LUT property blob size
is always a multiple of the LUT entry size. (stolen from ARM driver)
v3:
- Drop the LUT size check again since
drm_atomic_replace_property_blob_from_id() already covers this for
us. (Alexandru Gheorghe)
v4:
- Use an enum to describe possible test values rather than #define's;
this is cleaner to provide kerneldoc for. (Daniel Vetter)
- s/DRM_COLOR_LUT_INCREASING/DRM_COLOR_LUT_NON_DECREASING/. (Ville)
Cc: Uma Shankar <uma.shankar@intel.com>
Cc: Swati Sharma <swati2.sharma@intel.com>
Cc: Brian Starkey <Brian.Starkey@arm.com>
Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Brian Starkey <brian.starkey@arm.com>
Reviewed-by: Alexandru Gheorghe <alexandru-cosmin.gheorghe@arm.com>
Reviewed-by: Uma Shankar <uma.shankar@intel.com>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20181217224415.12848-1-matthew.d.roper@intel.com
When reading GEN11_GT_INTR_DWx closely after enabling the interrupts
in gen11_irq_postinstall, the returned value is garbage. This can
cause other parts of the setup code (e.g. gen11_reset_one_iir) to
think that there are interrupts to be cleared when there are none.
The garbage value is only seen on the first read done after the enable,
so this looks like a posting issue. Adding a posting read after enabling
the interrupts does indeed fix the problem.
Note that the posting read has been purposely added outside of
gen11_master_intr_enable since the issue has only been observed when the
full interrupt setup is performed.
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190123023227.8117-1-daniele.ceraolospurio@intel.com
The old debugfs fields was not following a naming partern and it was
a bit confusing.
So it went from:
~$ sudo more /sys/kernel/debug/dri/0/i915_edp_psr_status
Sink_Support: yes
PSR mode: PSR1
Enabled: yes
Busy frontbuffer bits: 0x000
Main link in standby mode: no
HW Enabled & Active bit: yes
Source PSR status: 0x24050006 [SRDONACK]
To:
~$ sudo more /sys/kernel/debug/dri/0/i915_edp_psr_status
Sink support: yes [0x03]
PSR mode: PSR1 enabled
Source PSR ctl: enabled [0x81f00e26]
Source PSR status: IDLE [0x04010006]
Busy frontbuffer bits: 0x00000000
The 'Main link in standby mode' was removed as it is not useful but
if needed by someone the information is still in the register value
of 'Source PSR ctl' inside of the brackets, PSR mode and Enabled was
squashed into PSR mode, some renames and reorders and we have this
cleaner version. This will also make easy to parse debugfs for IGT
tests.
v2: Printing sink PSR version with only 2 hex digits as it is a byte
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Suggested-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Reviewed-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Acked-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190117205548.28378-2-jose.souza@intel.com
