MTL's primary GT can continue to use the same engine TLB invalidation
programming as past Xe_HP-based platforms. However the media GT needs
some special handling:
* Invalidation registers on the media GT are singleton registers
(unlike the primary GT where they are still MCR).
* Since the GSC is now exposed as an engine, there's a new register to
use for TLB invalidation. The offset is identical to the compute
engine offset, but this is expected --- compute engines only exist on
the primary GT while the GSC only exists on the media GT.
* Although there's only a single GSC engine instance, it inexplicably
uses bit 1 to request invalidations rather than bit 0.
v2:
- Add a 'regs == xelpmp_regs' condition to the GSC instance handling.
If the registers change on a future platform, the GSC-specific
handling is likely to change as well. (Andrzej)
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Andrzej Hajda <andrzej.hajda@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Andrzej Hajda <andrzej.hajda@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230224012009.3594691-1-matthew.d.roper@intel.com
Although registers in the L3 bank/node configuration ranges are marked
as having "DEV" reset characteristics in the bspec, this appears to be a
hold-over from pre-Xe_HP platforms. In reality, these registers
maintain their values across engine resets, meaning that workarounds
and tuning settings targeting them should be placed on the GT
workaround list rather than an engine workaround list.
Note that an extra clue here is that these registers moved from the
RENDER forcewake domain to the GT forcewake domain in Xe_HP; generally
RCS/CCS engine resets should not lead to the reset of a register that
lives outside the RENDER domain.
Re-applying these registers on engine resets wouldn't actually hurt
anything, but is unnecessary and just makes it more confusing to anyone
trying to decipher how these registers really work.
v2:
- Also move DG2's Wa_14010648519 to the GT list. (Gustavo)
Cc: Gustavo Sousa <gustavo.sousa@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Gustavo Sousa <gustavo.sousa@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230209232228.859317-1-matthew.d.roper@intel.com
Register 0x9424 is not replicated on any platform, so it shouldn't be
declared with REG_MCR(). Declaring it with _MMIO() is basically
duplicate of the GEN7 version, so just remove the GEN8 and change all
the callers to use the right functions.
Old versions of the gen8 bspec page used to contain a table with MCR
registers, apparently implying 0x9400 - 0x94ff registers were
replicated. However that table went away and there is no information
related to the ranges for gen8 anymore. Moreover the current behavior of
the driver wouldn't do anything special for 0x9424 since there is no
equivalent table in intel_gt_mcr.c: the driver would just fallback to
intel_uncore_{read,write}(). Therefore, do not care about the possible
special case for gen8 and just use the register as non-MCR for all the
platforms.
One place doing read + write is also converted to intel_uncore_rmw().
v2: Reword commit message adding the justification wrt gen8
Fixes: a9e69428b1 ("drm/i915: Define MCR registers explicitly")
Cc: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Gustavo Sousa <gustavo.sousa@intel.com>
Cc: Matt Atwood <matthew.s.atwood@intel.com>
Cc: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230206165410.3056073-1-lucas.demarchi@intel.com
The UNSLICE_UNIT_LEVEL_CLKGATE register programmed by this workaround
has 'BUS' style reset, indicating that it does not lose its value on
engine resets. Furthermore, this register is part of the GT forcewake
domain rather than the RENDER domain, so it should not be impacted by
RCS engine resets. As such, we should implement this on the GT
workaround list rather than an engine list.
Bspec: 19219
Fixes: 3551ff9287 ("drm/i915/gen11: Moving WAs to rcs_engine_wa_init()")
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Gustavo Sousa <gustavo.sousa@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230201222831.608281-2-matthew.d.roper@intel.com
XEHPC_LNCFMISCCFGREG0 and XEHPC_L3SCRUB are both in MCR register ranges
on PVC (with HALFBSLICE and L3BANK replication respectively), so they
should be explicitly declared as MCR registers and use MCR-aware
workaround handlers.
The workarounds/tuning settings should still be applied properly on PVC
even without the MCR annotation, but readback verification on
CONFIG_DRM_I915_DEBUG_GEM builds could potentitally give false positive
"workaround lost on load" warnings on parts fused such that a unicast
read targets a terminated register instance.
Fixes: a9e69428b1 ("drm/i915: Define MCR registers explicitly")
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Gustavo Sousa <gustavo.sousa@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230201222831.608281-1-matthew.d.roper@intel.com
During module load the punit might still be busy with its booting
routines. During this time we try to communicate with it but we
fail because we don't receive any feedback from it and we return
immediately with a -EINVAL fatal error.
At this point the driver load is "dramatically" aborted. The
following error message notifies us about it.
i915 0000:4d:00.0: drm_WARN_ON_ONCE(timeout_base_ms > 3)
It would be enough to wait a little in order to give the punit
the chance to come up bright and shiny, ready to interact with
the driver.
Wait up 10 seconds for the punit to settle and complete any
outstanding transactions upon module load. If it still fails try
again with a longer timeout, 180s, 3 minutes. If it still fails
then return -EPROBE_DEFER, in order to give the punit a second
chance.
Even if these timers might look long, we should consider that the
punit, depending on the platforms, might need long times to
complete its routines. Besides we want to try anything possible
to move forward before deciding to abort the driver's load.
The issue has been reported in:
https://gitlab.freedesktop.org/drm/intel/-/issues/7814
The changes in this patch are valid only and uniquely during
boot. The common transactions with the punit during the driver's
normal operation are not affected.
Signed-off-by: Aravind Iddamsetty <aravind.iddamsetty@intel.com>
Co-developed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Andrzej Hajda <andrzej.hajda@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230206183236.109908-1-andi.shyti@linux.intel.com
DSM granularity is 1MB so make sure we stick to that.
The address set by firmware in GEN12_DSMBASE in driver initialization
doesn't mean "anything above that and until end of lmem is part of DSM".
In fact, there may be a few KB that is not part of DSM on the end of
lmem. How large is that space is platform-dependent, but since it's
always less than the DSM granularity, it can be simplified by simply
aligning the size down.
v2: replace "1 * SZ_1M" with SZ_1M (Andrzej).
v3: reword commit message to explain why the round down is needed
(Lucas)
Cc: Matthew Auld <matthew.auld@intel.com>
Suggested-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: Andrzej Hajda <andrzej.hajda@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230202180243.23637-1-nirmoy.das@intel.com
Engine resets are supposed to never fail. But in the case when one
does (due to unknown reasons that normally come down to a missing
w/a), it is useful to get as much information out of the system as
possible. Given that the GuC intentionally dies on such a situation,
it is not possible to get a guilty context notification back. So do a
manual search instead. Given that GuC is dead, this is safe because
GuC won't be changing the engine state asynchronously.
v2: Change comment to be less alarming (Tvrtko)
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230127002842.3169194-7-John.C.Harrison@Intel.com
There was a report of error captures occurring without any hung
context being indicated despite the capture being initiated by a 'hung
context notification' from GuC. The problem was not reproducible.
However, it is possible to happen if the context in question has no
active requests. For example, if the hang was in the context switch
itself then the breadcrumb write would have occurred and the KMD would
see an idle context.
In the interests of attempting to provide as much information as
possible about a hang, it seems wise to include the engine info
regardless of whether a request was found or not. As opposed to just
prentending there was no hang at all.
So update the error capture code to always record engine information
if a context is given. Which means updating record_context() to take a
context instead of a request (which it only ever used to find the
context anyway). And split the request agnostic parts of
intel_engine_coredump_add_request() out into a seaprate function.
v2: Remove a duplicate 'if' statement (Umesh) and fix a put of a null
pointer.
v3: Tidy up request locking code flow (Tvrtko)
v4: Pull in improved info message from next patch and fix up potential
leak of GuC register state (Daniele)
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com> (v2)
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230127002842.3169194-5-John.C.Harrison@Intel.com
When GuC support was added to error capture, the reference counting
around the request object was broken. Fix it up.
The context based search manages the spinlocking around the search
internally. So it needs to grab the reference count internally as
well. The execlist only request based search relies on external
locking, so it needs an external reference count but within the
spinlock not outside it.
The only other caller of the context based search is the code for
dumping engine state to debugfs. That code wasn't previously getting
an explicit reference at all as it does everything while holding the
execlist specific spinlock. So, that needs updaing as well as that
spinlock doesn't help when using GuC submission. Rather than trying to
conditionally get/put depending on submission model, just change it to
always do the get/put.
v2: Explicitly document adding an extra blank line in some dense code
(Andy Shevchenko). Fix multiple potential null pointer derefs in case
of no request found (some spotted by Tvrtko, but there was more!).
Also fix a leaked request in case of !started and another in
__guc_reset_context now that intel_context_find_active_request is
actually reference counting the returned request.
v3: Add a _get suffix to intel_context_find_active_request now that it
grabs a reference (Daniele).
v4: Split the intel_guc_find_hung_context change to a separate patch
and rename intel_context_find_active_request_get to
intel_context_get_active_request (Tvrtko).
v5: s/locking/reference counting/ in commit message (Tvrtko)
Fixes: dc0dad365c ("drm/i915/guc: Fix for error capture after full GPU reset with GuC")
Fixes: 573ba126ae ("drm/i915/guc: Capture error state on context reset")
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Andrzej Hajda <andrzej.hajda@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Cc: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Cc: Michael Cheng <michael.cheng@intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: Tejas Upadhyay <tejaskumarx.surendrakumar.upadhyay@intel.com>
Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: Aravind Iddamsetty <aravind.iddamsetty@intel.com>
Cc: Alan Previn <alan.previn.teres.alexis@intel.com>
Cc: Bruce Chang <yu.bruce.chang@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230127002842.3169194-3-John.C.Harrison@Intel.com
