Having two methods to wait on GT TLB invalidations is not ideal. Remove
xe_gt_tlb_invalidation_wait and only use GT TLB invalidation fences.
In addition to two methods being less than ideal, once GT TLB
invalidations are coalesced the seqno cannot be assigned during
xe_gt_tlb_invalidation_ggtt/range. Thus xe_gt_tlb_invalidation_wait
would not have a seqno to wait one. A fence however can be armed and
later signaled.
v3:
- Add explaination about coalescing to commit message
v4:
- Don't put dma fence if defined on stack (CI)
v5:
- Initialize ret to zero (CI)
v6:
- Use invalidation_fence_signal helper in tlb timeout (Matthew Auld)
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240719172905.1527927-3-matthew.brost@intel.com
Update PT layer so if a memory allocation for a PTE fails the error can
be propagated to the user without requiring the VM to be killed.
v5:
- change return value invalidation_fence_init to void (Matthew Auld)
v7:
- Invert i,j usage in two places (Matthew Auld)
- s/0/NULL (Matthew Auld)
- Don't ignore return value of xe_pt_new_shared (Matthew Auld)
- Don't check for NULL in xe_pt_entry (Matthew Auld)
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240704041652.272920-7-matthew.brost@intel.com
This aligns with the uAPI of an array of binds or single bind that
results in multiple GPUVA ops to be considered a single atomic
operations.
The design is roughly:
- xe_vma_ops is a list of xe_vma_op (GPUVA op)
- each xe_vma_op resolves to 0-3 PT ops
- xe_vma_ops creates a single job
- if at any point during binding a failure occurs, xe_vma_ops contains
the information necessary unwind the PT and VMA (GPUVA) state
v2:
- add missing dma-resv slot reservation (CI, testing)
v4:
- Fix TLB invalidation (Paulo)
- Add missing xe_sched_job_last_fence_add/test_dep check (Inspection)
v5:
- Invert i, j usage (Matthew Auld)
- Add helper to test and add job dep (Matthew Auld)
- Return on anything but -ETIME for cpu bind (Matthew Auld)
- Return -ENOBUFS if suballoc of BB fails due to size (Matthew Auld)
- s/do/Do (Matthew Auld)
- Add missing comma (Matthew Auld)
- Do not assign return value to xe_range_fence_insert (Matthew Auld)
v6:
- s/0x1ff/MAX_PTE_PER_SDI (Matthew Auld, CI)
- Check to large of SA in Xe to avoid triggering WARN (Matthew Auld)
- Fix checkpatch issues
v7:
- Rebase
- Support more than 510 PTEs updates in a bind job (Paulo, mesa testing)
v8:
- Rebase
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240704041652.272920-5-matthew.brost@intel.com
The default behavior of device atomics depends on the
VM type and buffer allocation types. Device atomics are
expected to function with all types of allocations for
traditional applications/APIs. Additionally, in compute/SVM
API scenarios with fault mode or LR mode VMs, device atomics
must work with single-region allocations. In all other cases
device atomics should be disabled by default also on platforms
where we know device atomics doesn't on work on particular
allocations types.
v3: fault mode requires LR mode so only check for LR mode
to determine compute API(Jose).
Handle SMEM+LMEM BO's migration to LMEM where device
atomics is expected to work. (Brian).
v2: Fix platform checks to correct atomics behaviour on PVC.
Acked-by: Michal Mrozek <michal.mrozek@intel.com>
Reviewed-by: Oak Zeng <oak.zeng@intel.com>
Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240430162529.21588-6-nirmoy.das@intel.com
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
The flags stored in the BO grew over time without following
much a naming pattern. First of all, get rid of the _BIT suffix that was
banned from everywhere else due to the guideline in
drivers/gpu/drm/i915/i915_reg.h that xe kind of follows:
Define bits using ``REG_BIT(N)``. Do **not** add ``_BIT`` suffix to the name.
Here the flags aren't for a register, but it's good practice to keep it
consistent.
Second divergence on names is the use or not of "CREATE". This is
because most of the flags are passed to xe_bo_create*() family of
functions, changing its behavior. However, since the flags are also
stored in the bo itself and checked elsewhere in the code, it seems
better to just omit the CREATE part.
With those 2 guidelines, all the flags are given the form
XE_BO_FLAG_<FLAG_NAME> with the following commands:
git grep -le "XE_BO_" -- drivers/gpu/drm/xe | xargs sed -i \
-e "s/XE_BO_\([_A-Z0-9]*\)_BIT/XE_BO_\1/g" \
-e 's/XE_BO_CREATE_/XE_BO_FLAG_/g'
git grep -le "XE_BO_" -- drivers/gpu/drm/xe | xargs sed -i -r \
-e 's/XE_BO_(DEFER_BACKING|SCANOUT|FIXED_PLACEMENT|PAGETABLE|NEEDS_CPU_ACCESS|NEEDS_UC|INTERNAL_TEST|INTERNAL_64K|GGTT_INVALIDATE)/XE_BO_FLAG_\1/g'
And then the defines in drivers/gpu/drm/xe/xe_bo.h are adjusted to
follow the coding style.
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240322142702.186529-3-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Rebinding might allocate page-table bos, causing evictions.
To support blocking locking during these evictions,
perform the rebinding in the drm_exec locking loop.
Also Reserve fence slots where actually needed rather than trying to
predict how many fence slots will be needed over a complete
wound-wait transaction.
v2:
- Remove a leftover call to xe_vm_rebind() (Matt Brost)
- Add a helper function xe_vm_validate_rebind() (Matt Brost)
v3:
- Add comments and squash with previous patch (Matt Brost)
Fixes: 24f947d58f ("drm/xe: Use DRM GPUVM helpers for external- and evicted objects")
Fixes: 29f424eb87 ("drm/xe/exec: move fence reservation")
Cc: Matthew Auld <matthew.auld@intel.com>
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/20240327091136.3271-5-thomas.hellstrom@linux.intel.com
For each rebind we insert a GuC TLB invalidation and add a
corresponding unordered TLB invalidation fence. This might
add a huge number of TLB invalidation fences to wait for so
rather than doing that, defer the TLB invalidation to the
next ring ops for each affected exec queue. Since the TLB
is invalidated on exec_queue switch, we need to invalidate
once for each affected exec_queue.
v2:
- Simplify if-statements around the tlb_flush_seqno.
(Matthew Brost)
- Add some comments and asserts.
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/20240327091136.3271-2-thomas.hellstrom@linux.intel.com
Currently scratch PTEs are write-enabled and points to a single scratch
page. This has the side effect that buggy applications with out-of-bounds
memory accesses may not notice the bad access since what's written may
be read back.
Instead use NULL PTEs as scratch PTEs. These always return 0 when reading,
and writing has no effect. As a slight benefit, we can also use huge NULL
PTEs.
One drawback pointed out is that debugging may be hampered since previously
when inspecting the content of the scratch page, it might be possible to
detect writes to out-of-bound addresses and possibly also
from where the out-of-bounds address originated. However since the scratch
page-table structure is kept, it will be easy to add back the single
RW-enabled scratch page under a debug define if needed.
Also update the kerneldoc accordingly and move the function to create the
scratch page-tables from xe_pt.c to xe_pt.h since it is accessing
vm structure internals and this also makes it possible to make it static.
v2:
- Don't try to encode scratch PTEs larger than 1GiB.
- Move xe_pt_create_scratch(), Update kerneldoc.
v3:
- Rebase.
Cc: Brian Welty <brian.welty@intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Acked-by: Lucas De Marchi <lucas.demarchi@intel.com> #for general direction.
Reviewed-by: Brian Welty <brian.welty@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231209151843.7903-3-thomas.hellstrom@linux.intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Allow userspace to directly control the pat_index for a given vm
binding. This should allow directly controlling the coherency, caching
behaviour, compression and potentially other stuff in the future for the
ppGTT binding.
The exact meaning behind the pat_index is very platform specific (see
BSpec or PRMs) but effectively maps to some predefined memory
attributes. From the KMD pov we only care about the coherency that is
provided by the pat_index, which falls into either NONE, 1WAY or 2WAY.
The vm_bind coherency mode for the given pat_index needs to be at least
1way coherent when using cpu_caching with DRM_XE_GEM_CPU_CACHING_WB. For
platforms that lack the explicit coherency mode attribute, we treat
UC/WT/WC as NONE and WB as AT_LEAST_1WAY.
For userptr mappings we lack a corresponding gem object, so the expected
coherency mode is instead implicit and must fall into either 1WAY or
2WAY. Trying to use NONE will be rejected by the kernel. For imported
dma-buf (from a different device) the coherency mode is also implicit
and must also be either 1WAY or 2WAY.
v2:
- Undefined coh_mode(pat_index) can now be treated as programmer
error. (Matt Roper)
- We now allow gem_create.coh_mode <= coh_mode(pat_index), rather than
having to match exactly. This ensures imported dma-buf can always
just use 1way (or even 2way), now that we also bundle 1way/2way into
at_least_1way. We still require 1way/2way for external dma-buf, but
the policy can now be the same for self-import, if desired.
- Use u16 for pat_index in uapi. u32 is massive overkill. (José)
- Move as much of the pat_index validation as we can into
vm_bind_ioctl_check_args. (José)
v3 (Matt Roper):
- Split the pte_encode() refactoring into separate patch.
v4:
- Rebase
v5:
- Check for and reject !coh_mode which would indicate hw reserved
pat_index on xe2.
v6:
- Rebase on removal of coh_mode from uapi. We just need to reject
cpu_caching=wb + pat_index with coh_none.
Testcase: igt@xe_pat
Bspec: 45101, 44235 #xe
Bspec: 70552, 71582, 59400 #xe2
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Pallavi Mishra <pallavi.mishra@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Cc: José Roberto de Souza <jose.souza@intel.com>
Cc: Filip Hazubski <filip.hazubski@intel.com>
Cc: Carl Zhang <carl.zhang@intel.com>
Cc: Effie Yu <effie.yu@intel.com>
Cc: Zhengguo Xu <zhengguo.xu@intel.com>
Cc: Francois Dugast <francois.dugast@intel.com>
Tested-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Acked-by: Zhengguo Xu <zhengguo.xu@intel.com>
Acked-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
The name "compute_mode" can be confusing since compute uses either this
mode or fault_mode to achieve the long-running semantics, and compute_mode
can, moving forward, enable fault_mode under the hood to work around
hardware limitations.
Also the name no_dma_fence_mode really refers to what we elsewhere call
long-running mode and the mode contrary to what its name suggests allows
dma-fences as in-fences.
So in an attempt to be more consistent, rename
no_dma_fence_mode -> lr_mode
compute_mode -> preempt_fence_mode
And adjust flags so that
preempt_fence_mode sets XE_VM_FLAG_LR_MODE
fault_mode sets XE_VM_FLAG_LR_MODE | XE_VM_FLAG_FAULT_MODE
v2:
- Fix a typo in the commit message (Oak Zeng)
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Oak Zeng <oak.zeng@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231127123349.23698-1-thomas.hellstrom@linux.intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
After noticing in logs there were still mentions to GEN6 registers, it
was clear commit d9b79ad275 ("drm/xe: Drop gen afixes from registers")
didn't take care of all the afixes. Some were added later, but there are
also constants and strings still using that. Continue the cleanup
removing the remaining ones.
To keep it consistent with code nearby, a few other changes are made:
- Remove prefix in INTEL_LEGACY_64B_CONTEXT
- Remove GEN8_CTX_L3LLC_COHERENT since it's unused
- Rename GEN9_FREQ_SCALER to GT_FREQUENCY_SCALER
v2: Use XELP_ as prefix for NUM_MOCS_ENTRIES and remove changes to
MOCS_ENTRIES as this is now done as part of a previous commit
(Matt Roper)
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20231117174049.527192-3-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
We're already using the half-open interval notation "[A, B)", that "-
1" there makes it wrong. Also, getting rid of the "-1" makes it much
easier to grep for the logs when you're looking for an address that's
the end of a vma and the start of another.
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Split functions that do only part of the pde/pte encoding and that can
be called by the different places. This normalizes how pde/pte are
encoded so they can be moved elsewhere in a subsequent change.
xe_pte_encode() was calling __pte_encode() with a NULL vma, which is the
opposite of what xe_pt_stage_bind_entry() does. Stop passing a NULL vma
and just split another function that deals with a vma rather than a bo.
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230927193902.2849159-2-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@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>
Integrated graphics 1270 and beyond should set the PTE_LM bit in the PTE
when it's stolen memory. Add a new function, xe_bo_is_stolen_devmem(),
and use it when encoding the PTE.
In some places in the spec the PTE bit is called "Local Memory",
abbreviated as LM, and in others it's called "Device Memory" (DM). Since
we moved away from "Local Memory" and preferred the "vram" terminology,
also rename the macros as DM to follow the name of the new function.
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://lore.kernel.org/r/20230726160708.3967790-7-lucas.demarchi@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Do not queue the rebind worker directly, rather use the helper
xe_vm_queue_rebind_worker. This ensures we use the correct work queue.
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
On MTL and beyond, the GPU performs non-coherent accesses to the PPGTT
page tables. These page tables should be mapped as CPU:WC.
Removes CAT errors triggered by xe_exec_basic@once-basic on MTL:
xe 0000:00:02.0: [drm:__xe_pt_bind_vma [xe]] Preparing bind, with range [1a0000...1a0fff) engine 0000000000000000.
xe 0000:00:02.0: [drm:xe_vm_dbg_print_entries [xe]] 1 entries to update
xe 0000:00:02.0: [drm:xe_vm_dbg_print_entries [xe]] 0: Update level 3 at (0 + 1) [0...8000000000) f:0
xe 0000:00:02.0: [drm] Engine memory cat error: guc_id=2
xe 0000:00:02.0: [drm] Engine memory cat error: guc_id=2
xe 0000:00:02.0: [drm] Timedout job: seqno=4294967169, guc_id=2, flags=0x4
v2:
- Rename to XE_BO_PAGETABLE to make it more clear that this BO is the
pagetable itself, rather than just being bound in the PPGTT. (Lucas)
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Acked-by: Nirmoy Das <nirmoy.das@intel.com>
Link: https://lore.kernel.org/r/20230725003433.1992137-3-matthew.d.roper@intel.com
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
