They can actually complete out-of-order, so allocate a unique
fence context for each fence.
Fixes: 5387e865d9 ("drm/xe: Add TLB invalidation fence after rebinds issued from execs")
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: <stable@vger.kernel.org> # v6.8+
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240327091136.3271-4-thomas.hellstrom@linux.intel.com
A force_wake_get failure means that the HW might not be awake for the
access we're doing; this can lead to an immediate error or it can be a
more subtle problem (e.g. a register read might return an incorrect
value that is still valid, leading the driver to make a wrong choice
instead of flagging an error).
We avoid an error from the force_wake function because callers might
handle or tolerate the error, but this only works if all callers
are checking the error code. The majority already do, but a few are not.
These are mainly falling into 3 categories, which are each handled
differently:
1) error capture: in this case we want to continue the capture, but we
log an info message in dmesg to notify the user that the capture
might have incorrect data.
2) ioctl: in this case we return a -EIO error to userspace
3) unabortable actions: these are scenarios where we can't simply abort
and retry and so it's better to just try it anyway because there is a
chance the HW is awake even with the failure. In this case we throw a
warning so we know there was a forcewake problem if something fails
down the line.
v2: use gt_WARN_ON where appropriate
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Tejas Upadhyay <tejas.upadhyay@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240318154924.3453513-1-daniele.ceraolospurio@intel.com
Forcewakes are not required for communication with the GuC via CTB
as it is a memory based interfaced. Acquring forcewakes takes
considerable time. With that, do not grab a forcewake when issuing a
GGTT TLB invalidation via the GuC.
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240229194520.200642-1-matthew.brost@intel.com
The xe_gt_tlb_invalidation layer implements TLB invalidations for a GuC
backend. Simply return if in execlists mode. A follow up may properly
implement the xe_gt_tlb_invalidation layer for both GuC and execlists.
Fixes: a9351846d9 ("drm/xe: Break of TLB invalidation into its own file")
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240222232021.3911545-4-matthew.brost@intel.com
We track TLB invalidation seqno per GT and we have GT oriented
message helpers, so it's better to use GT oriented log functions.
Cc: Matt Roper <matthew.d.roper@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://lore.kernel.org/r/20231218190629.502-3-michal.wajdeczko@intel.com
Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
On i915 we were adding new GuC ABI headers directly to guc_fwif.h
file since we were replacing old definitions from that file.
On xe driver we could do more and better by including ABI headers
only in files that need those definitions.
Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/741
Cc: Jani Nikula <jani.nikula@intel.com>
Acked-by: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://lore.kernel.org/r/20231128203203.1147-3-michal.wajdeczko@intel.com
Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Duplicating these helpers in almost every .c file is a bad idea.
Define them as inlines in .h file to allow proper reuse.
Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
The XE_WARN_ON macro maps to WARN_ON which is not justified
in many cases where only a simple debug check is needed.
Replace the use of the XE_WARN_ON macro with the new xe_assert
macros which relies on drm_*. This takes a struct drm_device
argument, which is one of the main changes in this commit. The
other main change is that the condition is reversed, as with
XE_WARN_ON a message is displayed if the condition is true,
whereas with xe_assert it is if the condition is false.
v2:
- Rebase
- Keep WARN splats in xe_wopcm.c (Matt Roper)
v3:
- Rebase
Signed-off-by: Francois Dugast <francois.dugast@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Replace calls to XE_BUG_ON() with calls XE_WARN_ON() which in turn calls
WARN() instead of BUG(). BUG() crashes the kernel and should only be
used when it is absolutely unavoidable in case of catastrophic and
unrecoverable failures, which is not the case here.
Signed-off-by: Francois Dugast <francois.dugast@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
In various test cases that put the system under a heavy load, we can
sometimes see errors with missed TLB invalidations. In such cases we see
the interrupt arrive for the invalidation from the GuC, however the
actual processing of the completion is pushed onto a workqueue and
handled with all the other CT stuff, which might take longer than
expected. Since we expect TLB invalidations to complete within a
reasonable amount of time (at most ~250ms), and they do seem pretty
critical, allow handling directly from the CT fast-path.
v2 (José):
- Actually use the correct spinlock/unlock_irq, since pending_lock is
grabbed from IRQ.
v3:
- Don't publish the TLB fence on the list until after we fully
initialize it and successfully do the CT send. The list is now only
protected by the spin_lock pending_lock and we can't hold that
across the entire TLB send operation.
v4 (Matt Brost):
- Be careful with racing against fast CT path writing the seqno,
before we have actually published the fence.
References: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/297
References: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/320
References: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/449
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
To help debugging, sample the current seqno_recv and dump it out if we
encounter a TLB timeout for the fences path.
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
We might have various kworkers waiting for TLB flushes to complete which
are not tracked with an explicit TLB fence, however at this stage that
will never happen since the CT is already disabled, so make sure we
signal them here under the assumption that we have completed a full GT
reset.
v2:
- We need to use seqno - 1 here. After acquiring ct->lock the seqno is
actually the next users seqno and not the pending one.
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
If we are in the middle of a GT reset or similar the CT might be
disabled, such that the CT send fails. However we already incremented
gt->tlb_invalidation.seqno which might lead to warnings, since we
effectively just skipped a seqno:
0000:00:02.0: drm_WARN_ON(expected_seqno != msg[0])
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Ensure we load gt->tlb_invalidation.seqno_recv once, and use that for
our seqno checking. The gt->tlb_invalidation_seqno_past is a shared
global variable and can potentially change at any point here. However
the checks here need to operate on a stable version of seqno_recv for
this to make any sense.
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
wake_up_all() and wait_event_timeout() already have the correct barriers
as per https://www.kernel.org/doc/Documentation/memory-barriers.txt.
This should ensure that the seqno_recv write can't be re-ordered wrt to
the actual wake_up_all() i.e we get woken up but there is no write. The
reader side with wait_event_timeout() also has the correct barriers.
With that drop the hand rolled smp_wmb(), which is anyway missing some
kind of matching barrier on the reader side.
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
This will help with the GPUVA port as the internals of struct xe_vma
will change.
v2: Update comment around helpers
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.kernel.org>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Rather than a backpointer to the xe_device, a GT should have a
backpointer to its tile (which can then be used to lookup the device if
necessary).
The gt_to_xe() helper macro (which moves from xe_gt.h to xe_gt_types.h)
can and should still be used to jump directly from an xe_gt to
xe_device.
v2:
- Fix kunit test build
- Move a couple changes to the previous patch. (Lucas)
Reviewed-by: Matt Atwood <matthew.s.atwood@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://lore.kernel.org/r/20230601215244.678611-4-matthew.d.roper@intel.com
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Lockdep is unhappy about ggtt->lock -> runtime_pm, where it seems
to think this can somehow get inverted. The ggtt->lock looks like a
potentially sensitive driver lock, so likely a sensible move to never
call the runtime_pm routines while holding it. Actually it looks like
d3cold wants to grab this, so perhaps this can indeed deadlock.
v2:
- Don't forget about xe_gt_tlb_invalidation_vma(), which now needs
explicit access_get.
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Seems to be a sensitive lock, where ct->lock looks to be primed with
fs_reclaim, so holding that and then allocating memory will cause
lockdep to complain. We need to change the ordering wrt to grabbing the
ct->lock and potentially grabbing the runtime_pm, since some of the
runtime_pm routines can allocate memory (or at least that's what lockdep
seems to suggest).
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Checking seqno_recv >= seqno looks like it will incorrectly report true
when the seqno has wrapped (not unlikely given
TLB_INVALIDATION_SEQNO_MAX). Calling xe_gt_tlb_invalidation_wait() might
then return before the flush has been completed by the GuC.
Fix this by treating a large negative delta as an indication that the
seqno has wrapped around. Similar to how we treat a large positive delta
as an indication that the seqno_recv must have wrapped around, but in
that case the seqno has likely also signalled.
It looks like we could also potentially make the seqno use the full
32bits as supported by the GuC.
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
It looks like when tlb_invalidation.seqno overflows
TLB_INVALIDATION_SEQNO_MAX, we start counting again from one, as per
send_tlb_invalidation(). This is also inline with initial value we give
it in xe_gt_tlb_invalidation_init(). When calculating the
expected_seqno we should also take this into account.
While we are here also print out the values if we ever trigger the
warning.
v2 (José):
- drm_WARN_ON() is preferred over plain WARN_ON(), since it gives
information on the originating device.
Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/248
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Sort includes and split them in blocks:
1) .h corresponding to the .c. Example: xe_bb.c should have a "#include
"xe_bb.h" first.
2) #include <linux/...>
3) #include <drm/...>
4) local includes
5) i915 includes
This is accomplished by running
`clang-format --style=file -i --sort-includes drivers/gpu/drm/xe/*.[ch]`
and ignoring all the changes after the includes. There are also some
manual tweaks to split the blocks.
v2: Also sort includes in headers
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
This means we are in the middle of a GT reset and no need to do TLB
invalidation so just signal invalidation fence immediately.
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Only the GuC should be issuing TLB invalidations if it is enabled. Part
of this patch is sanitize the device on driver unload to ensure we do
not send GuC based TLB invalidations during driver unload.
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
If the platform supports range based TLB invalidations use them. Hide
these details in the xe_gt_tlb_invalidation layer.
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Endless fences are not good, add a TDR to cleanup any invalidation
fences which have not received an invalidation message within a timeout
period.
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
This will help debug issues with TLB invalidation fences.
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
This gets tricky as we can't do the TLB invalidation until the unbind
operation is done on the hardware and we can't signal the unbind as
complete until the TLB invalidation is done. To work around this we
create an unbind fence which does a TLB invalidation after unbind is
done on the hardware, signals on TLB invalidation completion, and this
fence is installed in the BO dma-resv slot and installed in out-syncs
for the unbind operation.
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Suggested-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com
Suggested-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Fence will be signaled when TLB invalidation completion.
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Suggested-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
TLB invalidations no longer just restricted to USM, move the variables
to own sub-structure.
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
TLB invalidation is used by more than USM (page faults) so break this
code out into its own file.
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
