Replace the calls to ttm_eu_reserve_buffers() by using the drm_exec
helper instead. Also make sure the locking loop covers any calls to
xe_bo_validate() / ttm_bo_validate() so that these function calls may
easily benefit from being called from within an unsealed locking
transaction and may thus perform blocking dma_resv locks in the future.
For the unlock we remove an assert that the vm->rebind_list is empty
when locks are released. Since if the error path is hit with a partly
locked list, that assert may no longer hold true we chose to remove it.
v3:
- Don't accept duplicate bo locks in the rebind worker.
v5:
- Loop over drm_exec objects in reverse when unlocking.
v6:
- We can't keep the WW ticket when retrying validation on OOM. Fix.
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/20230908091716.36984-5-thomas.hellstrom@linux.intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
If multiple bind ops in an array of binds touch the same address range
invalid GPUVA operations are generated as each GPUVA operation is
generated based on the orignal GPUVA state. To fix this, after each
GPUVA operations is generated, commit the GPUVA operation updating the
GPUVA state so subsequent bind ops can see a current GPUVA state.
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Remap ops have 3 parts: unmap, prev, and next. The commit step can fail
on any of these. Add a flag for each to these so the unwind is only done
the steps that have been committed.
v2: (Rodrigo) Use bit macros
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Starting with Xe2, a 5-level page table is always used, regardless of
the actual virtual address range supported by the platform. The two
values need to be tracked separately in the device descriptor since Xe2
platforms only have a 48 bit virtual address range.
Bspec: 59505, 65637, 70817
Cc: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
If we are closing a vm, mark each vma as XE_VMA_DESTROYED and skip
touching the rebind_list if this is seen on the eviction path. That way
we can safely drop the vm dma-resv lock on the close path without
needing to worry about racing with the eviction path trying to add stuff
to the rebind_list which can corrupt our contended list, since the
destroy and rebind links are the same list entry underneath.
References: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/514
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
To workaround a HW bug on DG2, driver is required to map the whole
ppgtt virtual address space before GPU workload submission. Thus
set the XE_VM_FLAG_SCRATCH_PAGE flag during vm create so the whole
address space is mapped to point to scratch page.
v1:
- Move the workaround implementation from xe_vm_create to
xe_vm_create_ioctl - Brian
- Reorder error checking in xe_vm_create_ioctl - Jose
- Implement WA only for DG2-G10 and DG2-G12
Signed-off-by: Oak Zeng <oak.zeng@intel.com>
Reviewed-by: Brian Welty <brian.welty@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
When a rebind is skipped, we must set the max pte size of the newly
created vma to value of the old vma as we do not pte walk for the new
vma. Without this future rebinds may be incorrectly skipped due to the
wrong max pte size. Null binds are more likely to expose this bug as
larger ptes are more frequently used compared to normal bindings.
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Testcase: dEQP-VK.sparse_resources.buffer.ssbo.sparse_residency.buffer_size_2_24
Reported-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Fixes: 8f33b4f054 ("drm/xe: Avoid doing rebinds")
Reference: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23045
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>
We must use migrate engine for page fault binds in order to avoid a
deadlock as the migrate engine has a reserved BCS instance which cannot
be stuck on a fault. To use the migrate engine the engine argument to
xe_migrate_update_pgtables must be NULL, this was incorrectly wired up
so vm->eng[tile_id] was always being used. Fix this.
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Combine the userptr, rebind, and destroy links into a union as
the lists these links belong to are mutually exclusive.
v2: Adjust which lists are combined (Thomas H)
v3: Add kernel doc why this is safe (Thomas H), remove related change
of list_del_init -> list_del (Rodrigo)
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
If we dont change page sizes we can avoid doing rebinds rather just do a
partial unbind. The algorithm to determine its page size is greedy as we
assume all pages in the removed VMA are the largest page used in the
VMA.
v2: Don't exceed 100 lines
v3: struct xe_vma_op_unmap remove in different patch, remove XXX comment
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
We currently have a race between bind engines which can result in
corrupted page tables leading to faults.
A simple example:
bind A 0x0000-0x1000, engine A, has unsatisfied in-fence
bind B 0x1000-0x2000, engine B, no in-fences
exec A uses 0x1000-0x2000
Bind B will pass bind A and exec A will fault. This occurs as bind A
programs the root of the page table in a bind job which is held up by an
in-fence. Bind B in this case just programs a leaf entry of the
structure.
To fix use range-fence utility to track cross bind engine conflicts. In
the above example bind A would insert an dependency into the range-fence
tree with a key of 0x0-0x7fffffffff, bind B would find that dependency
and its bind job would scheduled behind the unsatisfied in-fence and
bind A's job.
Reviewed-by: Maarten Lankhorst<maarten.lankhorst@linux.intel.com>
Co-developed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reduce the number of warnings reported by checkpatch.pl from 118 to 48 by
addressing those warnings types:
LEADING_SPACE
LINE_SPACING
BRACES
TRAILING_SEMICOLON
CONSTANT_COMPARISON
BLOCK_COMMENT_STYLE
RETURN_VOID
ONE_SEMICOLON
SUSPECT_CODE_INDENT
LINE_CONTINUATIONS
UNNECESSARY_ELSE
UNSPECIFIED_INT
UNNECESSARY_INT
MISORDERED_TYPE
Signed-off-by: Francois Dugast <francois.dugast@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Rather than open coding VM binds and VMA tracking, use the GPUVA
library. GPUVA provides a common infrastructure for VM binds to use mmap
/ munmap semantics and support for VK sparse bindings.
The concepts are:
1) xe_vm inherits from drm_gpuva_manager
2) xe_vma inherits from drm_gpuva
3) xe_vma_op inherits from drm_gpuva_op
4) VM bind operations (MAP, UNMAP, PREFETCH, UNMAP_ALL) call into the
GPUVA code to generate an VMA operations list which is parsed, committed,
and executed.
v2 (CI): Add break after default in case statement.
v3: Rebase
v4: Fix some error handling
v5: Use unlocked version VMA in error paths
v6: Rebase, address some review feedback mainly Thomas H
v7: Fix compile error in xe_vma_op_unwind, address checkpatch
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
We cannot recover a VM if a rebind worker hits an error, ban the VM if
happens to ensure we do not attempt to place this VM on the hardware
again.
A follow up will inform the user if this happens.
v2: Return -ECANCELED in exec VM closed or banned, check for closed or
banned within VM lock.
v3: Fix lockdep splat by looking engine outside of vm->lock
v4: Fix error path when engine lookup fails
v5: Add debug message in rebind worker on error, update comments wrt
locking, add xe_vm_close helper
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
xe_vm_create used the IOCTL create flags in a few places rather than the
internal VM flags and this just happened to work as these values
matched. This is risky (and incorrect) as the internal flag values are
free to change. Fix this and use the internal VM flag values.
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Use the TTM LRU bulk move for BOs tied to a VM. Update the bulk moves
LRU position on every exec.
v2: Bulk move for compute VMs, use WARN rather than BUG
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
We only need to try to lock a BO if it's external as non-external BOs
share the dma-resv with the already locked VM. Trying to lock
non-external BOs caused an issue (list corruption) in an uncoming patch
which adds bulk LRU move. Since this code isn't needed, remove it.
v2: New commit message, s/mattthew/matthew/
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Add uAPI and implementation for NULL bindings. A NULL binding is defined
as writes dropped and read zero. A single bit in the uAPI has been added
which results in a single bit in the PTEs being set.
NULL bindings are intendedd to be used to implement VK sparse bindings,
in particular residencyNonResidentStrict property.
v2: Fix BUG_ON shown in VK testing, fix check patch warning, fix
xe_pt_scan_64K, update __gen8_pte_encode to understand NULL bindings,
remove else if vma_addr
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Suggested-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Using uninitialized variables leads to undefined behavior.
Moreover, it causes the compiler to complain with:
../drivers/gpu/drm/xe/xe_vm.c:3265:40: error: variable 'vma' is uninitialized when used here [-Werror,-Wuninitialized]
../drivers/gpu/drm/xe/xe_rtp.c:118:36: error: variable 'i' is uninitialized when used here [-Werror,-Wuninitialized]
../drivers/gpu/drm/xe/xe_mocs.c:449:3: error: variable 'flags' is uninitialized when used here [-Werror,-Wuninitialized]
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230523135020.345596-1-michal@hardline.pl
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
This define is for internal PTE flags rather than fields in the hardware
PTEs, rename as such. This will help in an upcoming patch to avoid
further confusion.
Reviewed-by: Francois Dugast <francois.dugast@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
A mix of the system unbound wq and Xe ordered wq was used for the
rebind, only use the Xe ordered wq. This will ensure only 1 rebind is
occuring at a time providing a somewhat clunky work around for short
comings in TTM wrt to memory contention. Once the TTM memory contention
is resolved we should be able to use a dedicated non-ordered workqueue.
Also add helper to queue rebind worker to avoid using wrong workqueue
going forward.
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
For scratch table mode we need to cover the case where a scratch PTE might
have been pre-fetched and cached and used instead of that of the newly
bound vma.
For compute vms, invalidate TLB globally using GuC before signalling
bind complete. For !long-running vms, invalidate TLB at batch start.
Also document how TLB invalidation works.
v2:
- Fix a pointer to the comment about TLB invalidation (Jose Souza).
- Add a bool to the vm whether we want to invalidate TLB at batch start.
- Invalidate TLB also on BCS- and video engines at batch start where
needed.
- Use BIT() macro instead of explicit shift.
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Tested-by: José Roberto de Souza <jose.souza@intel.com> #v1
Reported-by: José Roberto de Souza <jose.souza@intel.com> #v1
Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/291
Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/291
Acked-by: 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>
Migration primarily focuses on the memory associated with a tile, so it
makes more sense to track this at the tile level (especially since the
driver was already skipping migration operations on media GTs).
Note that the blitter engine used to perform the migration always lives
in the tile's primary GT today. In theory that could change if media
GTs ever start including blitter engines in the future, but we can
extend the design if/when that happens in the future.
v2:
- Fix kunit test build
- Kerneldoc parameter name update
v3:
- Removed leftover prototype for removed function. (Gustavo)
- Remove unrelated / unwanted error handling change. (Gustavo)
Cc: Gustavo Sousa <gustavo.sousa@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Acked-by: Gustavo Sousa <gustavo.sousa@intel.com>
Link: https://lore.kernel.org/r/20230601215244.678611-15-matthew.d.roper@intel.com
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Since memory and address spaces are a tile concept rather than a GT
concept, we need to plumb tile-based handling through lots of
memory-related code.
Note that one remaining shortcoming here that will need to be addressed
before media GT support can be re-enabled is that although the address
space is shared between a tile's GTs, each GT caches the PTEs
independently in their own TLB and thus TLB invalidation should be
handled at the GT level.
v2:
- Fix kunit test build.
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://lore.kernel.org/r/20230601215244.678611-13-matthew.d.roper@intel.com
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Create a new xe_tile structure to begin separating the concept of "tile"
from "GT." A tile is effectively a complete GPU, and a GT is just one
part of that. On platforms like MTL, there's only a single full GPU
(tile) which has its IP blocks provided by two GTs. In contrast, a
"multi-tile" platform like PVC is basically multiple complete GPUs
packed behind a single PCI device.
For now, just create xe_tile as a simple wrapper around xe_gt. The
items in xe_gt that are truly tied to the tile rather than the GT will
be moved in future patches. Support for multiple GTs per tile (i.e.,
the MTL standalone media case) will also be re-introduced in a future
patch.
v2:
- Fix kunit test build
- Move hunk from next patch to use local tile variable rather than
direct xe->tiles[id] accesses. (Lucas)
- Mention compute in kerneldoc. (Rodrigo)
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://lore.kernel.org/r/20230601215244.678611-3-matthew.d.roper@intel.com
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
It looks like the driver only wants to track one vma for each external
object per vm. However it looks like bo_has_vm_references_locked() will
ignore any vma that is marked as vma->destroyed (not actually destroyed
yet). If we then mark our externally tracked vma as destroyed and then
create a new vma for the same object and vm, we can have two externally
tracked vma for the same object and vm. When the destroy actually
happens it tries to move the external tracking to a different vma, but
in this case it is already being tracked, leading to double list add
errors. It should be safe to simply drop the destroyed check in
bo_has_vm_references(), since the actual destroy will switch the
external tracking to the next available vma.
Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/290
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
