The GuC firmware had defined the interface for Translation Look-Aside
Buffer (TLB) invalidation. We should use this interface when
invalidating the engine and GuC TLBs.
Add additional functionality to intel_gt_invalidate_tlb, invalidating
the GuC TLBs and falling back to GT invalidation when the GuC is
disabled.
The invalidation is done by sending a request directly to the GuC
tlb_lookup that invalidates the table. The invalidation is submitted as
a wait request and is performed in the CT event handler. This means we
cannot perform this TLB invalidation path if the CT is not enabled.
If the request isn't fulfilled in two seconds, this would constitute
an error in the invalidation as that would constitute either a lost
request or a severe GuC overload.
With this new invalidation routine, we can perform GuC-based GGTT
invalidations. GuC-based GGTT invalidation is incompatible with
MMIO invalidation so we should not perform MMIO invalidation when
GuC-based GGTT invalidation is expected.
The additional complexity incurred in this patch will be necessary for
range-based tlb invalidations, which will be platformed in the future.
Signed-off-by: Prathap Kumar Valsan <prathap.kumar.valsan@intel.com>
Signed-off-by: Bruce Chang <yu.bruce.chang@intel.com>
Signed-off-by: Chris Wilson <chris.p.wilson@intel.com>
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Signed-off-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Signed-off-by: Aravind Iddamsetty <aravind.iddamsetty@intel.com>
Signed-off-by: Fei Yang <fei.yang@intel.com>
CC: Andi Shyti <andi.shyti@linux.intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Acked-by: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231017180806.3054290-4-jonathan.cavitt@intel.com
If we can't find a free fence register to handle a fault in the GMADR
range just return VM_FAULT_NOPAGE without populating the PTE so that
userspace will retry the access and trigger another fault. Eventually
we should find a free fence and the fault will get properly handled.
A further improvement idea might be to reserve a fence (or one per CPU?)
for the express purpose of handling faults without having to retry. But
that would require some additional work.
Looks like this may have gotten broken originally by
commit 39965b3766 ("drm/i915: don't trash the gtt when running out of fences")
as that changed the errno to -EDEADLK which wasn't handle by the gtt
fault code either. But later in commit 2feeb52859 ("drm/i915/gt: Fix
-EDEADLK handling regression") I changed it again to -ENOBUFS as -EDEADLK
was now getting used for the ww mutex dance. So this fix only makes
sense after that last commit.
Cc: stable@vger.kernel.org
Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/9479
Fixes: 2feeb52859 ("drm/i915/gt: Fix -EDEADLK handling regression")
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231012132801.16292-1-ville.syrjala@linux.intel.com
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
If an active context has been banned (e.g. Ctrl+C killed) then it is
likely to be reset as part of evicting it from the hardware. That
results in a 'ignoring context reset notification: banned = 1'
message at info level. This confuses/concerns people and makes them
think something has gone wrong when it hasn't.
There is already a debug level message with essentially the same
information. So drop the 'ignore' info level one and just add the
'ignore' flag to the debug level one instead (which will therefore not
appear by default but will still show up in CI runs).
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230921182033.135448-1-John.C.Harrison@Intel.com
Chaining user engines happens in multiple passes during driver
initialization, mutating its type along the way. It starts off with a
simple lock-less linked list (struct llist_node/head) populated by
intel_engine_add_user() which later gets sorted and converted to an
intermediate regular list (struct list_head) just to be converted once
more to its final rb-tree structure (struct rb_node/root) in
intel_engines_driver_register().
All of these types overlay the uabi_node/uabi_engines members which is
unfortunate but safe if one takes care about using the rb-tree based
structure only after the conversion has completed. However, mistakes
happen and commit 1ec23ed712 ("drm/i915: Use uabi engines for the
default engine map") violated that assumption, as the multiple type
evolution was all to easy hidden behind casts papering over it.
Make the type evolution of uabi_node/uabi_engines more visible by
putting all members into an anonymous union and use the correctly typed
member in its various users. This allows us to drop quite some ugly
casts and, hopefully, make the evolution of the members better
recognisable to avoid future mistakes.
Signed-off-by: Mathias Krause <minipli@grsecurity.net>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230928182019.10256-3-minipli@grsecurity.net
Commit 1ec23ed712 ("drm/i915: Use uabi engines for the default engine
map") switched from using for_each_engine() to for_each_uabi_engine() to
iterate over the user engines. While this seems to be a sensible change,
it's only safe to do when the engines are actually chained using the
rb-tree structure which is not the case during early driver
initialization where it can be either a lock-less list or regular
double-linked list.
In fact, the modesetting initialization code may end up calling
default_engines() through the fb helper code while the engines list
is still llist_node-based:
i915_driver_probe() ->
intel_display_driver_probe() ->
intel_fbdev_init() ->
drm_fb_helper_init() ->
drm_client_init() ->
drm_client_open() ->
drm_file_alloc() ->
i915_driver_open() ->
i915_gem_open() ->
i915_gem_context_open() ->
i915_gem_create_context() ->
default_engines()
Using for_each_uabi_engine() in default_engines() is therefore wrong, as
it would try to interpret the llist as rb-tree, making it find no engine
at all, as the rb_left and rb_right members will still be NULL, as they
haven't been initialized yet.
To fix this type confusion register the engines earlier and at the same
time reduce the amount of code that has to deal with the intermediate
llist state.
Reported-by: sanity checks in grsecurity
Suggested-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Fixes: 1ec23ed712 ("drm/i915: Use uabi engines for the default engine map")
Signed-off-by: Mathias Krause <minipli@grsecurity.net>
Cc: Jonathan Cavitt <jonathan.cavitt@intel.com>
Cc: 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/20230928182019.10256-2-minipli@grsecurity.net
[tursulin: fixed commit tag typo]
Toggle binder context ready status when needed.
To issue gpu commands, the driver must be primed to receive
requests. Maintain binder-based GGTT update disablement until driver
probing completes. Moreover, implement a temporary disablement
of blitter prior to entering suspend, followed by re-enablement
post-resume. This is acceptable as those transition periods are
mostly single threaded.
v2: move changes to lower levels from i915_driver.c(Jani).
use new function for setting context ready status.
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
Signed-off-by: Oak Zeng <oak.zeng@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230926083742.14740-7-nirmoy.das@intel.com
Implement GGTT update method with blitter command, MI_UPDATE_GTT
and install those handlers if a platform requires that.
v2: Make sure we hold the GT wakeref and Blitter engine wakeref before
we call mutex_lock/intel_context_enter below. When GT/engine are not
awake, the intel_context_enter calls into some runtime pm function which
can end up with kmalloc/fs_reclaim. But trigger fs_reclaim holding a
mutex lock is not allowed because shrinker can also try to hold the same
mutex lock. It is a circular lock. So hold the GT/blitter engine wakeref
before calling mutex_lock, to fix the circular lock.
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
Signed-off-by: Oak Zeng <oak.zeng@intel.com>
Acked-by: Oak Zeng <oak.zeng@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230926083742.14740-6-nirmoy.das@intel.com
When runtime pm is first woken, it will synchronously call the
registered callbacks for the device. These callbacks
may pull in their own forest of locks, which we do not want to
conflate with the intel_wakeref.mutex. A second minor benefit to
reducing the coverage of the mutex, is that it will reduce
contention for frequent sleeps and wakes (such as when being used
for soft-rc6).
v2: remove usage of fetch_and_zero() and other improvements(Jani)
Signed-off-by: Chris Wilson <chris.p.wilson@intel.com>
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230926083742.14740-2-nirmoy.das@intel.com
During resume, the steer semaphore on GT1 was observed to be held. The
hardware team has confirmed the safety of clearing steer semaphores
for all GTs during driver load/resume, as no lock acquisitions can occur
in this process by other agents.
v2: reset on resume not in intel_gt_init().
v3: do the reset on intel_gt_resume_early()
v4: do general sanitization for all GTs(Matt)
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230928130015.6758-3-nirmoy.das@intel.com
Commit ade8a0f598 ("drm/i915: Make all GPU resets atomic") added a
preempt disable section over the hardware reset callback to prepare the
driver for being able to reset from atomic contexts.
In retrospect I can see that the work item at a time was about removing
the struct mutex from the reset path. Code base also briefly entertained
the idea of doing the reset under stop_machine in order to serialize
userspace mmap and temporary glitch in the fence registers (see
eb8d0f5af4 ("drm/i915: Remove GPU reset dependence on struct_mutex"),
but that never materialized and was soon removed in 2caffbf117
("drm/i915: Revoke mmaps and prevent access to fence registers across
reset") and replaced with a SRCU based solution.
As such, as far as I can see, today we still have a requirement that
resets must not sleep (invoked from submission tasklets), but no need to
support invoking them from a truly atomic context.
Given that the preemption section is problematic on RT kernels, since the
uncore lock becomes a sleeping lock and so is invalid in such section,
lets try and remove it. Potential downside is that our short waits on GPU
to complete the reset may get extended if CPU scheduling interferes, but
in practice that probably isn't a deal breaker.
In terms of mechanics, since the preemption disabled block is being
removed we just need to replace a few of the wait_for_atomic macros into
busy looping versions which will work (and not complain) when called from
non-atomic sections.
v2:
* Fix timeouts which are now in us. (Andi)
* Update one comment as a drive by. (Andi)
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Chris Wilson <chris.p.wilson@intel.com>
Cc: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Andi Shyti <andi.shyti@linux.intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230926100855.61722-1-tvrtko.ursulin@linux.intel.com
Ideally the busyness worker should take a gt pm wakeref because the
worker only needs to be active while gt is awake. However, the gt_park
path cancels the worker synchronously and this complicates the flow if
the worker is also running at the same time. The cancel waits for the
worker and when the worker releases the wakeref, that would call gt_park
and would lead to a deadlock.
The resolution is to take the global pm wakeref if runtime pm is already
active. If not, we don't need to update the busyness stats as the stats
would already be updated when the gt was parked.
Note:
- We do not requeue the worker if we cannot take a reference to runtime
pm since intel_guc_busyness_unpark would requeue the worker in the
resume path.
- If the gt was parked longer than time taken for GT timestamp to roll
over, we ignore those rollovers since we don't care about tracking the
exact GT time. We only care about roll overs when the gt is active and
running workloads.
- There is a window of time between gt_park and runtime suspend, where
the worker may run. This is acceptable since the worker will not find
any new data to update busyness.
v2: (Daniele)
- Edit commit message and code comment
- Use runtime pm in the worker
- Put runtime pm after enabling the worker
- Use Link tag and add Fixes tag
v3: (Daniele)
- Reword commit and comments and add details
Link: https://gitlab.freedesktop.org/drm/intel/-/issues/7077
Fixes: 77cdd054dd ("drm/i915/pmu: Connect engine busyness stats from GuC to pmu")
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230925192117.2497058-1-umesh.nerlige.ramappa@intel.com
There is no reason to add gtt_offset to the cached head/tail pointers
stream->oa_buffer.head and stream->oa_buffer.tail. This causes the code to
constantly add gtt_offset and subtract gtt_offset and is error
prone.
It is much simpler to maintain stream->oa_buffer.head and
stream->oa_buffer.tail without adding gtt_offset to them and just allow for
the gtt_offset when reading/writing from/to HW registers.
v2: Minor tweak to commit message due to dropping patch in previous series
Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230920040211.2351279-1-ashutosh.dixit@intel.com
Invalidate instruction and State cache bit using INDIRECT_CTX on
every gpu context switch for gen12.
The goal of this workaround is to actually perform an explicit
invalidation of that cache (by re-writing the register) during every GPU
context switch, which is accomplished via a "workaround batchbuffer"
that's attached to the context via INDIRECT_CTX. (Matt Roper)
Please refer [1] for more reviews and comment on the same patch
[1] https://patchwork.freedesktop.org/series/123377/
v2:
- Remove extra parentheses from the condition (Lucas)
- Align spacing and new line (Lucas)
v3:
- Fix commit message.
v4:
- Only Gen12 changes are kept and Remove DG2+ condition (Matt Roper)
- Fix the commit message for r-b (Matt Roper)
- Rename the register bit in define
v5:
- Move out this workaround from golden context init (Matt Roper)
- Use INDIRECT_CTX to set bit on each GPU context switch (Matt Roper)
v6:
- Change IP Version base condition for Gen12 (Matt Roper)
- Made imperative form of commit version messages (Suraj)
- s/Added/Add in patch header (Suraj)
v7:
- In version descriptions s/Ropper/Roper (Matt Atwood)
BSpec: 11354
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Cc: Suraj Kandpal <suraj.kandpal@intel.com>
Cc: Matt Atwood <matthew.s.atwood@intel.com>
Signed-off-by: Dnyaneshwar Bhadane <dnyaneshwar.bhadane@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/20230914202000.1069884-1-dnyaneshwar.bhadane@intel.com
