Matt Roper [Fri, 5 Sep 2025 21:56:15 +0000 (14:56 -0700)]
drm/xe: Never report L3 bank mask for media GT going forward
We currently report an L3 bank mask as part of the GT topology on both
GTs (primary and media) because a copy of the L3 bank fuse register
exists on both GTs (e.g., $gsi_offset + 0x9130 on Xe3). After recent
discussions it's come to light that the only known userspace software
that uses this part of the uapi (the compute UMD and Mesa) only uses the
value reported for the primary GT; the value reported for the media GT
is ignored by both projects, and the media UMDs don't have any use for
L3 information today. Since we always strive to have our uapi match the
specific needs of userspace and not include additional unused baggage,
let's officially drop L3 bank reporting on the media GT going forward
and only keep it around for the primary GT where it actually gets used.
This change will only apply to future platforms (Xe3 and later); even
though it would probably be safe to remove it from Xe1/Xe2 as well, we
don't want to take any chances with changing existing ABI.
Note that we'd already disabled reading/reporting of the L3 bank for the
media GT on PTL in commit 9ab440a9d042 ("drm/xe/ptl: L3bank mask is not
available on the media GT") because it was discovered that the copy of
the fuse registers on the media GT were just reporting a bogus ~0 value
rather than an accurate mask. So this is just extending that PTL
behavior forward to WCL and other future platforms. Note that we're
also free to reinstate this part of the uapi in the future if/when some
new userspace consumer emerges that _does_ have a use for media-specific
L3 bank masks.
drm/xe/debugfs: Don't expose dgfx residencies attributes on VF
In addition of checking if we are running on the BATTLEMAGE,
we should also check for not being a VF driver, as VFs can't
access necessary registers, and doing so leads to:
| .. [drm] GT0: VF is trying to read an inaccessible register 0x35b004+0x0
| RIP: 0010:xe_gt_sriov_vf_read32+0x5e2/0x8a0 [xe]
| Call Trace:
| xe_mmio_read32+0x110/0x280 [xe]
| read_residency_counter+0x42/0xd0 [xe]
| dgfx_pkg_residencies_show+0x115/0x190 [xe]
| .. [drm] Package G2 counter failed to read, ret -19
or
| .. [drm] GT0: VF is trying to read an inaccessible register 0x35b004+0x0
| RIP: 0010:xe_gt_sriov_vf_read32+0x5e2/0x8a0 [xe]
| Call Trace:
| xe_mmio_read32+0x110/0x280 [xe]
| read_residency_counter+0x42/0xd0 [xe]
| dgfx_pcie_link_residencies_show+0xe7/0x160 [xe]
| .. [drm] PCIE LINK L0 RESIDENCY counter failed to read, ret -19
Similarly, there is no point to expose inject_csc_hw_error on VFs,
as HW errors support is already disabled for VFs.
We only need single set of VF CCS contexts, they are not per-tile
as initial implementation might suggest. Move all VF CCS data from
xe_tile.sriov.vf to xe_device.sriov.vf. Also rename some structs to
align with the usage and fix their kernel-doc.
This will allow as to drop ugly IS_VF_CCS_BB_VALID macro.
Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com> Cc: Satyanarayana K V P <satyanarayana.k.v.p@intel.com> Cc: Matthew Brost <matthew.brost@intel.com> Reviewed-by: Matthew Brost <matthew.brost@intel.com> Reviewed-by: Satyanarayana K V P <satyanarayana.k.v.p@intel.com> Link: https://lore.kernel.org/r/20250908123025.747-6-michal.wajdeczko@intel.com
drm/xe/vf: Use single check when calling VF CCS functions
All xe_sriov_vf_ccs() functions but init() expect to be called
when initialization was successful and CCS handling is ready.
Update IS_VF_CCS_READY macro and use it as single entry guard.
Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com> Cc: Satyanarayana K V P <satyanarayana.k.v.p@intel.com> Cc: Matthew Brost <matthew.brost@intel.com> Reviewed-by: Satyanarayana K V P <satyanarayana.k.v.p@intel.com> Link: https://lore.kernel.org/r/20250908123025.747-5-michal.wajdeczko@intel.com
We only use this macro once and we can open-code it to explicitly
show relevant conditions and avoid duplications.
Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com> Cc: Satyanarayana K V P <satyanarayana.k.v.p@intel.com> Reviewed-by: Matthew Brost <matthew.brost@intel.com> Reviewed-by: Satyanarayana K V P <satyanarayana.k.v.p@intel.com> Link: https://lore.kernel.org/r/20250908123025.747-4-michal.wajdeczko@intel.com
This function is dedicated for use by the VFs, we shouldn't use
name that might suggests it's general purpose. While there, update
asserts to better reflect intended usage.
Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com> Cc: Satyanarayana K V P <satyanarayana.k.v.p@intel.com> Cc: Matthew Brost <matthew.brost@intel.com> Reviewed-by: Satyanarayana K V P <satyanarayana.k.v.p@intel.com> Link: https://lore.kernel.org/r/20250908123025.747-2-michal.wajdeczko@intel.com
Lucas De Marchi [Fri, 5 Sep 2025 16:22:37 +0000 (09:22 -0700)]
drm/xe/configfs: Use config_group_put()
configfs has a config_group_put() helper that was adopted by
commit 88df7939d728 ("drm/xe/configfs: Rename struct xe_config_device").
Another pending work to add psmi later landed in commit afe902848b41 ("drm/xe/configfs: Allow to enable PSMI") and didn't use
the helper.
Use config_group_put() consistently to hide the inner workings of
configfs. No change in behavior since it does exactly the same thing
as currently being done.
John Harrison [Thu, 4 Sep 2025 19:57:51 +0000 (12:57 -0700)]
drm/xe/guc: Fix badly worded error message
If a GuC id lookup failed, the error message was 'Not engine present',
which is bad in multiple ways - incorrect English and 'engines' are
now called 'exec queues' in this context. So fix it.
drm/xe: Block exec and rebind worker while evicting for suspend / hibernate
When the xe pm_notifier evicts for suspend / hibernate, there might be
racing tasks trying to re-validate again. This can lead to suspend taking
excessive time or get stuck in a live-lock. This behaviour becomes
much worse with the fix that actually makes re-validation bring back
bos to VRAM rather than letting them remain in TT.
Prevent that by having exec and the rebind worker waiting for a completion
that is set to block by the pm_notifier before suspend and is signaled
by the pm_notifier after resume / wakeup.
It's probably still possible to craft malicious applications that block
suspending. More work is pending to fix that.
v3:
- Avoid wait_for_completion() in the kernel worker since it could
potentially cause work item flushes from freezable processes to
wait forever. Instead terminate the rebind workers if needed and
re-launch at resume. (Matt Auld)
v4:
- Fix some bad naming and leftover debug printouts.
- Fix kerneldoc.
- Use drmm_mutex_init() for the xe->rebind_resume_lock (Matt Auld).
- Rework the interface of xe_vm_rebind_resume_worker (Matt Auld).
drm/xe: Attempt to bring bos back to VRAM after eviction
VRAM+TT bos that are evicted from VRAM to TT may remain in
TT also after a revalidation following eviction or suspend.
This manifests itself as applications becoming sluggish
after buffer objects get evicted or after a resume from
suspend or hibernation.
If the bo supports placement in both VRAM and TT, and
we are on DGFX, mark the TT placement as fallback. This means
that it is tried only after VRAM + eviction.
This flaw has probably been present since the xe module was
upstreamed but use a Fixes: commit below where backporting is
likely to be simple. For earlier versions we need to open-
code the fallback algorithm in the driver.
v2:
- Remove check for dgfx. (Matthew Auld)
- Update the xe_dma_buf kunit test for the new strategy (CI)
- Allow dma-buf to pin in current placement (CI)
- Make xe_bo_validate() for pinned bos a NOP.
Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/5995 Fixes: a78a8da51b36 ("drm/ttm: replace busy placement with flags v6") Cc: Matthew Brost <matthew.brost@intel.com> Cc: Matthew Auld <matthew.auld@intel.com> Cc: <stable@vger.kernel.org> # v6.9+ Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com> Reviewed-by: Matthew Auld <matthew.auld@intel.com> Link: https://lore.kernel.org/r/20250904160715.2613-2-thomas.hellstrom@linux.intel.com
drm/xe/migrate: Remove unneeded emit_pte() when copying CCS only
In xe_migrate_copy(), when copy_only_ccs is true, we only need two
emit_pte() calls one for the BO and one for the raw CCS storage.
However, the current implementation issues three emit_pte() calls,
resulting in an unnecessary PTE programming job.
This fix removes the redundant emit_pte() call to avoid programming
the same PTEs twice and reducing overhead during CCS-only migration.
v2: Preserve correct behavior on DG2, which requires both CCS and
page copies.
drm/xe: Fix broken kernel-doc for the struct xe_bo
Use correct multi-line kernel-doc style if required.
Some members were described only in the commit message.
Some other members were described using wrong names.
Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com> Cc: Matthew Brost <matthew.brost@intel.com> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com> Cc: Satyanarayana K V P <satyanarayana.k.v.p@intel.com> Reviewed-by: Satyanarayana K V P <satyanarayana.k.v.p@intel.com> Link: https://lore.kernel.org/r/20250904144026.7222-1-michal.wajdeczko@intel.com
Michal Wajdeczko [Fri, 29 Aug 2025 17:19:21 +0000 (19:19 +0200)]
drm/xe/kunit: Drop xe_wa_test_exit
Remove xe_wa_test_exit() as it could crach the KUnit kernel in
case of hitting some asserts in xe_wa_test_init() as test->priv
could not be pointing to expected data.
| # xe_wa_gt: ASSERTION FAILED at drivers/gpu/drm/xe/tests/xe_wa_test.c:34
| Expected ret == 0, but
| ret == -19 (0xffffffffffffffed)
|Bus error - the host /dev/shm or /tmp mount likely just ran out of space
|Kernel panic - not syncing: Kernel mode signal 7
Note that there is no need to call drm_kunit_helper_free_device()
since our fake device allocated by drm_kunit_helper_alloc_device()
will be cleaned up automatically.
Michal Wajdeczko [Fri, 29 Aug 2025 17:19:19 +0000 (19:19 +0200)]
drm/xe/kunit: Drop custom struct platform_test_case
Custom struct platform_test_case definition in xe_wa is now almost
identical to generic struct xe_pci_fake_data defintiion except the
.name member, which could be generated by xe_pci_fake_data_desc().
Michal Wajdeczko [Fri, 29 Aug 2025 17:19:18 +0000 (19:19 +0200)]
drm/xe/kunit: Introduce xe_pci_fake_data_desc()
We already use struct xe_pci_fake_data to provide custom config of
the fake PCI device and soon we will be using this struct also as
direct parameter for the parameterized Xe KUnit tests.
Add function to generate description based on that config data.
For platform or subplatform name lookup pciidlist which already
have definitions of all supported platforms.
Matthew Auld [Fri, 29 Aug 2025 16:47:16 +0000 (17:47 +0100)]
drm/xe: improve dma-resv handling for backup object
Since the dma-resv is shared we don't need to reserve and add a fence
slot fence twice, plus no need to loop through the dependencies.
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: Jonathan Cavitt <jonathan.cavitt@intel.com> Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com> Link: https://lore.kernel.org/r/20250829164715.720735-2-matthew.auld@intel.com
Matthew Auld [Thu, 28 Aug 2025 14:24:39 +0000 (15:24 +0100)]
drm/xe/pt: unify xe_pt_svm_pre_commit with userptr
We now use the same notifier lock for SVM and userptr, with that we can
combine xe_pt_userptr_pre_commit and xe_pt_svm_pre_commit.
v2: (Matt B)
- Re-use xe_svm_notifier_lock/unlock for userptr.
- Combine svm/userptr handling further down into op_check_svm_userptr.
v3:
- Only hide the ops if we lack DRM_GPUSVM, since we also need them for
userptr.
Suggested-by: Matthew Brost <matthew.brost@intel.com> Signed-off-by: Matthew Auld <matthew.auld@intel.com> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com> Reviewed-by: Matthew Brost <matthew.brost@intel.com> Link: https://lore.kernel.org/r/20250828142430.615826-18-matthew.auld@intel.com
Matthew Auld [Thu, 28 Aug 2025 14:24:38 +0000 (15:24 +0100)]
drm/xe/userptr: replace xe_hmm with gpusvm
Goal here is cut over to gpusvm and remove xe_hmm, relying instead on
common code. The core facilities we need are get_pages(), unmap_pages()
and free_pages() for a given useptr range, plus a vm level notifier
lock, which is now provided by gpusvm.
v2:
- Reuse the same SVM vm struct we use for full SVM, that way we can
use the same lock (Matt B & Himal)
v3:
- Re-use svm_init/fini for userptr.
v4:
- Allow building xe without userptr if we are missing DRM_GPUSVM
config. (Matt B)
- Always make .read_only match xe_vma_read_only() for the ctx. (Dafna)
v5:
- Fix missing conversion with CONFIG_DRM_XE_USERPTR_INVAL_INJECT
v6:
- Convert the new user in xe_vm_madise.
Signed-off-by: Matthew Auld <matthew.auld@intel.com> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com> Cc: Dafna Hirschfeld <dafna.hirschfeld@intel.com> Cc: Matthew Brost <matthew.brost@intel.com> Reviewed-by: Matthew Brost <matthew.brost@intel.com> Link: https://lore.kernel.org/r/20250828142430.615826-17-matthew.auld@intel.com
Matthew Auld [Thu, 28 Aug 2025 14:24:37 +0000 (15:24 +0100)]
drm/xe/vm: split userptr bits into separate file
This will simplify compiling out the bits that depend on DRM_GPUSVM in a
later patch. Without this we end up littering the code with ifdef
checks, plus it becomes hard to be sure that something won't blow at
runtime due to something not being initialised, even though it passed
the build. Should be no functional change here.
Matthew Auld [Thu, 28 Aug 2025 14:24:35 +0000 (15:24 +0100)]
drm/gpusvm: refactor core API to use pages struct
Refactor the core API of get/unmap/free pages to all operate on
drm_gpusvm_pages. In the next patch we want to export a simplified core
API without needing fully blown svm range etc.
Suggested-by: Matthew Brost <matthew.brost@intel.com> Signed-off-by: Matthew Auld <matthew.auld@intel.com> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com> Reviewed-by: Matthew Brost <matthew.brost@intel.com> Link: https://lore.kernel.org/r/20250828142430.615826-14-matthew.auld@intel.com
Matthew Auld [Thu, 28 Aug 2025 14:24:34 +0000 (15:24 +0100)]
drm/gpusvm: pull out drm_gpusvm_pages substructure
Pull the pages stuff from the svm range into its own substructure, with
the idea of having the main pages related routines, like get_pages(),
unmap_pages() and free_pages() all operating on some lower level
structures, which can then be re-used for stuff like userptr.
v2:
- Move seq into pages struct (Matt B)
v3:
- Small kernel-doc fixes
Suggested-by: Matthew Brost <matthew.brost@intel.com> Signed-off-by: Matthew Auld <matthew.auld@intel.com> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com> Reviewed-by: Matthew Brost <matthew.brost@intel.com> Link: https://lore.kernel.org/r/20250828142430.615826-13-matthew.auld@intel.com
Matthew Auld [Thu, 28 Aug 2025 14:24:33 +0000 (15:24 +0100)]
drm/gpusvm: use more selective dma dir in get_pages()
If we are only reading the memory then from the device pov the direction
can be DMA_TO_DEVICE. This aligns with the xe-userptr code. Using the
most restrictive data direction to represent the access is normally a
good idea.
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: Thomas Hellström <thomas.hellstrom@linux.intel.com> Reviewed-by: Matthew Brost <matthew.brost@intel.com> Link: https://lore.kernel.org/r/20250828142430.615826-12-matthew.auld@intel.com
Matthew Auld [Thu, 28 Aug 2025 14:24:32 +0000 (15:24 +0100)]
drm/gpusvm: fix hmm_pfn_to_map_order() usage
Handle the case where the hmm range partially covers a huge page (like
2M), otherwise we can potentially end up doing something nasty like
mapping memory which is outside the range, and maybe not even mapped by
the mm. Fix is based on the xe userptr code, which in a future patch
will directly use gpusvm, so needs alignment here.
v2:
- Add kernel-doc (Matt B)
- s/fls/ilog2/ (Thomas)
Reported-by: Thomas Hellström <thomas.hellstrom@linux.intel.com> Signed-off-by: Matthew Auld <matthew.auld@intel.com> Cc: Matthew Brost <matthew.brost@intel.com> Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com> Link: https://lore.kernel.org/r/20250828142430.615826-11-matthew.auld@intel.com
Add Wa_18041344222 for Xe2_HPG that requires disabling
the perf mode for subslice count for eustall sampling
when the enabled slices are discontiguous.
drm/xe/mcr: Make xe_gt_mcr_get_dss_steering() input gt a const
Make gt, input parameter to xe_gt_mcr_get_dss_steering(), a
constant. This would allow xe_gt_mcr_get_dss_steering() to
be called from functions that have gt as const to struct xe_gt.
drm/xe/configfs: Don't expose survivability_mode if not applicable
The survivability_mode attribute is applicable only for DGFX and
platforms newer than BATTLEMAGE. Use .is_visible() hook to hide
this attribute when above conditions are not met. Remove code that
was trying to fix such configuration during the runtime.
Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com> Cc: Lucas De Marchi <lucas.demarchi@intel.com> Cc: Riana Tauro <riana.tauro@intel.com> Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com> Reviewed-by: Stuart Summers <stuart.summers@intel.com> Link: https://lore.kernel.org/r/20250902131744.5076-4-michal.wajdeczko@intel.com
drm/xe/configfs: Prepare to filter-out configfs attributes
Implement empty ops.is_visible hook to allow filtering-out any not
supported attributes, as not all of them are applicable on all xe
platforms.
Since during creation of each new configfs directory we are looking
for xe device descriptor to validate that xe driver supports given
PCI device, store reference to that descriptor to allow later use
while doing attribute filtering.
drm/xe/configfs: Don't touch survivability_mode on fini
This is a user controlled configfs attribute, we should not
modify that outside the configfs attr.store() implementation.
Fixes: bc417e54e24b ("drm/xe: Enable configfs support for survivability mode") Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com> Cc: Lucas De Marchi <lucas.demarchi@intel.com> Cc: Riana Tauro <riana.tauro@intel.com> Reviewed-by: Stuart Summers <stuart.summers@intel.com> Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com> Link: https://lore.kernel.org/r/20250904103521.7130-1-michal.wajdeczko@intel.com
drm/xe/guc: Set upper limit of H2G retries over CTB
The GuC communication protocol allows GuC to send NO_RESPONSE_RETRY
reply message to indicate that due to some interim condition it can
not handle incoming H2G request and the host shall resend it.
But in some cases, due to errors, this unsatisfied condition might
be final and this could lead to endless retries as it was recently
seen on the CI:
To avoid such dangerous loops allow only limited number of retries
(for now 50) and add some delays (n * 5ms) to slow down the rate of
resending this repeated request.
Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com> Cc: John Harrison <John.C.Harrison@Intel.com> Cc: Matthew Brost <matthew.brost@intel.com> Reviewed-by: Stuart Summers <stuart.summers@intel.com> Reviewed-by: Julia Filipchuk <julia.filipchuk@intel.com> Link: https://lore.kernel.org/r/20250903223330.6408-1-michal.wajdeczko@intel.com
Michal Wajdeczko [Fri, 29 Aug 2025 20:11:03 +0000 (22:11 +0200)]
drm/xe/debugfs: Move sa_info from gt to tile directory
Our drm-based suballocator is implemented per-tile so it is better
to show its debug information also per-tile debugfs directory, not
under per-gt directory as it is done today.
To allow adding more per-tile attributes, prepare necessary helper
functions, like we already did for per-gt or per-uc attributes.
drm/xe/vm: Fix error handling in xe_vm_query_vmas_attrs_ioctl()
copy_to_user() returns the number of bytes not copied on failure, not a
negative error code. Update the logic to return -EFAULT instead of the
number of bytes to correctly signal the error.
Fixes: 418807860e94 ("drm/xe/uapi: Add UAPI for querying VMA count and memory attributes") Cc: Matthew Brost <matthew.brost@intel.com> Cc: Dan Carpenter <dan.carpenter@linaro.org> Reviewed-by: Tejas Upadhyay <tejas.upadhyay@intel.com> Link: https://lore.kernel.org/r/20250828104933.3839825-3-himal.prasad.ghimiray@intel.com Signed-off-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Matthew Brost [Fri, 29 Aug 2025 17:22:32 +0000 (10:22 -0700)]
drm/xe: Add more SVM GT stats
Add more SVM GT stats which give visibility to where time is spent in
the SVM page fault handler. Stats include number of faults at a given
size, total SVM page fault time, migration time in us, copy time in us,
copy kb, get pages time in us, and bind time in us. Will help in tuning
SVM for performance.
v2:
- Include local changes
v3:
- Add tlb invalidation + valid page fault + per size copy size stats
v4:
- Ensure gt not NULL when incrementing SVM copy stats
- Normalize stats names
- Use magic macros to generate increment functions for ranges
v7:
- Use DEF_STAT_STR (Michal)
Matthew Brost [Fri, 29 Aug 2025 17:22:31 +0000 (10:22 -0700)]
drm/xe: Add clearing stats to GT debugfs
It helpful to clear GT stats, run a test cases which is being profiled,
and look at the results of the stats from the individual test case. Make
stats entry writable and upon write clear the stats.
v5:
- Drop clear_stats debugfs entry (Lucas)
v6:
- Use xe_gt_stats_clear rather than helper (Michal)
- Rework loop in xe_gt_stats_clear (Michal)
WA 22021007897 should also be applied to Graphics Versions 30.00, 30.01
and 30.03. To make it simple, simply use the range [3000, 3003] that
should be ok as there isn't a 3002 and if it's added, the WA list would
need to be revisited anyway.
Cc: Matt Atwood <matthew.s.atwood@intel.com> Cc: Gustavo Sousa <gustavo.sousa@intel.com> Signed-off-by: Tangudu Tilak Tirumalesh <tilak.tirumalesh.tangudu@intel.com> Reviewed-by: Matt Atwood <matthew.s.atwood@intel.com> Link: https://lore.kernel.org/r/20250827-wa-22021007897-v1-1-96922eb52af4@intel.com Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
drm/xe/vf: Enable CCS save/restore only on supported GUC versions
CCS save/restore is supported starting with GuC 70.48.0 (compatibility
version 1.23.0). Gate the feature on the GuC firmware version and keep it
disabled on older or unsupported versions.
Fixes: f3009272ff2e ("drm/xe/vf: Create contexts for CCS read write") Signed-off-by: Satyanarayana K V P <satyanarayana.k.v.p@intel.com> Cc: Michal Wajdeczko <michal.wajdeczko@intel.com> Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com> Cc: Andi Shyti <andi.shyti@kernel.org> Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com> Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com> Link: https://lore.kernel.org/r/20250902103256.21658-2-satyanarayana.k.v.p@intel.com
drm/xe/guc: Add devm release action to safely tear down CT
When a buffer object (BO) is allocated with the XE_BO_FLAG_GGTT_INVALIDATE
flag, the driver initiates TLB invalidation requests via the CTB mechanism
while releasing the BO. However a premature release of the CTB BO can lead
to system crashes, as observed in:
Introduce a devm-managed release action during xe_guc_ct_init() and
xe_guc_ct_init_post_hwconfig() to ensure proper CTB disablement before
resource deallocation, preventing the use-after-free scenario.
Signed-off-by: Satyanarayana K V P <satyanarayana.k.v.p@intel.com> Cc: Michal Wajdeczko <michal.wajdeczko@intel.com> Cc: Matthew Brost <matthew.brost@intel.com> Cc: Matthew Auld <matthew.auld@intel.com> Cc: Summers Stuart <stuart.summers@intel.com> Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com> Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com> Link: https://lore.kernel.org/r/20250901072541.31461-1-satyanarayana.k.v.p@intel.com
Thomas Hellström [Thu, 28 Aug 2025 13:48:37 +0000 (15:48 +0200)]
drm/xe: Fix incorrect migration of backed-up object to VRAM
If an object is backed up to shmem it is incorrectly identified
as not having valid data by the move code. This means moving
to VRAM skips the -EMULTIHOP step and the bo is cleared. This
causes all sorts of weird behaviour on DGFX if an already evicted
object is targeted by the shrinker.
Fix this by using ttm_tt_is_swapped() to identify backed-up
objects.
Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/5996 Fixes: 00c8efc3180f ("drm/xe: Add a shrinker for xe bos") Cc: Matthew Brost <matthew.brost@intel.com> Cc: Matthew Auld <matthew.auld@intel.com> Cc: <stable@vger.kernel.org> # v6.15+ Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com> Reviewed-by: Matthew Auld <matthew.auld@intel.com> Link: https://lore.kernel.org/r/20250828134837.5709-1-thomas.hellstrom@linux.intel.com
drm/xe/uapi: Fix kernel-doc formatting for madvise and vma_query
Correct kernel-doc formatting issues in the UAPI definitions for
madvise and VMA query interfaces to resolve docutils warnings during
documentation build.
Fixes: 418807860e94 ("drm/xe/uapi: Add UAPI for querying VMA count and memory attributes") Fixes: 231bb0ee7aa5 ("drm/xe/uapi: Add madvise interface") Cc: Matthew Brost <matthew.brost@intel.com> Cc: Lucas De Marchi <lucas.demarchi@intel.com> Signed-off-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com> Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com> Link: https://lore.kernel.org/r/20250828071516.3838110-1-himal.prasad.ghimiray@intel.com Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Riana Tauro [Mon, 25 Aug 2025 10:35:37 +0000 (16:05 +0530)]
drm/xe/nvm: Use root tile mmio
To allow initialization of nvm during early probe for future usecases,
use root tile instead of root gt to access mmios, as gt is not
yet initialized at early probe.
drm/xe/tests: Make cross-device dma-buf BOs CPU-visible on small BAR
Small-BAR systems (e.g., SR-IOV VFs in VMs) expose only a subset of
VRAM via PCI/BAR. Exporting a BO outside that window fails, and the
selftests also do CPU fill/verify.
Set XE_BO_FLAG_NEEDS_CPU_ACCESS for cross-device variants to force
CPU-mappable placement and keep tests reliable. Large-BAR/P2P setups
are unaffected.
Suggested-by: Matthew Auld <matthew.auld@intel.com> Signed-off-by: Marcin Bernatowicz <marcin.bernatowicz@linux.intel.com> Cc: Matthew Auld <matthew.auld@intel.com> Reviewed-by: Matthew Auld <matthew.auld@intel.com> Link: https://lore.kernel.org/r/20250814145950.430231-1-marcin.bernatowicz@linux.intel.com Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
Matthew Auld [Fri, 8 Aug 2025 11:04:53 +0000 (12:04 +0100)]
drm/xe/migrate: make MI_TLB_INVALIDATE conditional
When clearing VRAM we should be able to skip invalidating the TLBs if we
are only using the identity map to access VRAM (which is the common
case), since no modifications are made to PTEs on the fly. Also since we
use huge 1G entries within the identity map, there should be a pretty
decent chance that the next packet(s) (if also clears) can avoid a tree
walk if we don't shoot down the TLBs, like if we have to process a long
stream of clears.
For normal moves/copies, we usually always end up with the src or dst
being system memory, meaning we can't only rely on the identity map and
will also need to emit PTEs and so will always require a TLB flush.
v2:
- Update commit to explain the situation for normal copies (Matt B)
- Rebase on latest changes
Signed-off-by: Matthew Auld <matthew.auld@intel.com> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@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> Link: https://lore.kernel.org/r/20250808110452.467513-2-matthew.auld@intel.com
Matthew Brost [Tue, 26 Aug 2025 18:29:11 +0000 (18:29 +0000)]
drm/xe: Split TLB invalidation code in frontend and backend
The frontend exposes an API to the driver to send invalidations, handles
sequence number assignment, synchronization (fences), and provides a
timeout mechanism. The backend issues the actual invalidation to the
hardware (or firmware).
The new layering easily allows issuing TLB invalidations to different
hardware or firmware interfaces.
Matthew Brost [Tue, 26 Aug 2025 18:29:09 +0000 (18:29 +0000)]
drm/xe: Prep TLB invalidation fence before sending
It is a bit backwards to add a TLB invalidation fence to the pending
list after issuing the invalidation. Perform this step before issuing
the TLB invalidation in a helper function.
v2: Make sure the seqno_lock mutex covers the send as well (Matt)
Matthew Brost [Tue, 26 Aug 2025 18:29:08 +0000 (18:29 +0000)]
drm/xe: Decouple TLB invalidations from GT
Decouple TLB invalidations from the GT by updating the TLB invalidation
layer to accept a `struct xe_tlb_inval` instead of a `struct xe_gt`.
Also, rename *gt_tlb* to *tlb*. The internals of the TLB invalidation
code still operate on a GT, but this is now hidden from the rest of the
driver.
Matthew Brost [Tue, 26 Aug 2025 18:29:06 +0000 (18:29 +0000)]
drm/xe: Add xe_tlb_inval structure
Extract TLB invalidation state into a structure to decouple TLB
invalidations from the GT, allowing the structure to be embedded
anywhere in the driver.
Stuart Summers [Tue, 26 Aug 2025 18:29:04 +0000 (18:29 +0000)]
drm/xe: Cancel pending TLB inval workers on teardown
Add a new _fini() routine on the GT TLB invalidation
side to handle this worker cleanup on driver teardown.
v2: Move the TLB teardown to the gt fini() routine called during
gt_init rather than in gt_alloc. This way the GT structure stays
alive for while we reset the TLB state.
Stuart Summers [Tue, 26 Aug 2025 18:29:03 +0000 (18:29 +0000)]
drm/xe: Move explicit CT lock in TLB invalidation sequence
Currently the CT lock is used to cover TLB invalidation
sequence number updates. In an effort to separate the GuC
back end tracking of communication with the firmware from
the front end TLB sequence number tracking, add a new lock
here to specifically track those sequence number updates
coming in from the user.
Apart from the CT lock, we also have a pending lock to
cover both pending fences and sequence numbers received
from the back end. Those cover interrupt cases and so
it makes not to overload those with sequence numbers
coming in from new transactions. In that way, we'll employ
a mutex here.
v2: Actually add the correct lock rather than just dropping
it... (Matt)
Lucas De Marchi [Tue, 26 Aug 2025 15:32:11 +0000 (08:32 -0700)]
drm/xe/configfs: Block runtime attribute changes
Although it's possible to change the attributes in runtime, they have no
effect after the driver is already bound to the device. Check for that
and return -EBUSY in that case.
This should help users understand what's going on when the behavior is
not changing even if the value from the configfs is "right", but it got
to that state too late.
Xin Wang [Wed, 27 Aug 2025 00:06:33 +0000 (17:06 -0700)]
drm/xe: Ensure GT is in C0 during resumes
This patch ensures the gt will be awake for the entire duration
of the resume sequences until GuCRC takes over and GT-C6 gets
re-enabled.
Before suspending GT-C6 is kept enabled, but upon resume, GuCRC
is not yet alive to properly control the exits and some cases of
instability and corruption related to GT-C6 can be observed.
The commit 9ab440a9d042 ("drm/xe/ptl: L3bank mask is not
available on the media GT") added a workaround to ignore
the fuse register that L3 bank availability as it did not
contain valid values. Same is true for WCL therefore extend
the workaround to cover it.
Add support to handle CSC firmware reported errors. When CSC firmware
errors are encoutered, a error interrupt is received by the GFX device as
a MSI interrupt.
Device Source control registers indicates the source of the error as CSC
The HEC error status register indicates that the error is firmware reported
Depending on the type of error, the error cause is written to the HEC
Firmware error register.
On encountering such CSC firmware errors, the graphics device is
non-recoverable from driver context. The only way to recover from these
errors is firmware flash.
System admin/userspace is notified of the necessity of firmware flash
with a combination of vendor-specific drm device edged uevent, dmesg logs
and runtime survivability sysfs. It is the responsiblity of the consumer
to verify all the actions and then trigger a firmware flash using tools
like fwupd.
$ udevadm monitor --property --kernel
monitor will print the received events for:
KERNEL - the kernel uevent
Riana Tauro [Tue, 26 Aug 2025 06:34:15 +0000 (12:04 +0530)]
drm/xe: Add support to handle hardware errors
Gfx device reports two classes of errors: uncorrectable and
correctable. Depending on the severity uncorrectable errors are further
classified Non-Fatal and Fatal.
Correctable and Non-Fatal errors: These errors are reported as MSI. Bits in
the Master Interrupt Register indicate the class of the error.
The source of the error is then read from the Device Error Source
Register.
Fatal errors: These are reported as PCIe errors
When a PCIe error is asserted, the OS will perform a SBR (Secondary
Bus reset) which causes the driver to reload. The error registers are
sticky and the values are maintained through SBR.
Add basic support to handle these errors.
Bspec: 50875, 53073, 53074, 53075, 53076
v2: Format commit message (Umesh)
v3: fix documentation (Stuart)
Riana Tauro [Tue, 26 Aug 2025 06:34:13 +0000 (12:04 +0530)]
drm/xe/xe_survivability: Add support for Runtime survivability mode
Certain runtime firmware errors can cause the device to be in a unusable
state requiring a firmware flash to restore normal operation.
Runtime Survivability Mode indicates firmware flash is necessary by
wedging the device and exposing survivability mode sysfs.
The below sysfs is an indication that device is in survivability mode
/sys/bus/pci/devices/<device>/survivability_mode
v2: Fix kernel-doc (Umesh)
v3: Add user friendly dmesg (Frank)
Riana Tauro [Tue, 26 Aug 2025 06:34:11 +0000 (12:04 +0530)]
drm/xe: Add a helper function to set recovery method
Add a helper function to set recovery method. The recovery
method can be set before declaring the device wedged and sending the
drm wedged uevent. If no method is set, default unbind/re-bind method
will be set.
Riana Tauro [Tue, 26 Aug 2025 06:34:09 +0000 (12:04 +0530)]
drm: Add a vendor-specific recovery method to drm device wedged uevent
Address the need for a recovery method (firmware flash on Firmware errors)
introduced in the later patches of Xe KMD.
Whenever XE KMD detects a firmware error, a firmware flash is required to
recover the device to normal operation.
The initial proposal to use 'firmware-flash' as a recovery method was
not applicable to other drivers and could cause multiple recovery
methods specific to vendors to be added.
To address this a more generic 'vendor-specific' method is introduced,
guiding users to refer to vendor specific documentation and system logs
for detailed vendor specific recovery procedure.
Add a recovery method 'WEDGED=vendor-specific' for such errors.
Vendors must provide additional recovery documentation if this method
is used.
It is the responsibility of the consumer to refer to the correct vendor
specific documentation and usecase before attempting a recovery.
For example: If driver is XE KMD, the consumer must refer
to the documentation of 'Device Wedging' under 'Documentation/gpu/xe/'.
v2: fix documentation (Raag)
v3: add more details to commit message (Sima, Rodrigo, Raag)
add an example script to the documentation (Raag)
v4: use consistent naming (Raag)
v5: fix commit message
v6: add more documentation
Cc: André Almeida <andrealmeid@igalia.com> Cc: Christian König <christian.koenig@amd.com> Cc: David Airlie <airlied@gmail.com> Cc: Simona Vetter <simona.vetter@ffwll.ch> Cc: Maxime Ripard <mripard@kernel.org> Signed-off-by: Riana Tauro <riana.tauro@intel.com> Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com> Acked-by: Maxime Ripard <mripard@kernel.org> Link: https://lore.kernel.org/r/20250826063419.3022216-3-riana.tauro@intel.com Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
drm/xe/uapi: Add UAPI for querying VMA count and memory attributes
Introduce the DRM_IOCTL_XE_VM_QUERY_MEMORY_RANGE_ATTRS ioctl to allow
userspace to query memory attributes of VMAs within a user specified
virtual address range.
Userspace first calls the ioctl with num_mem_ranges = 0,
sizeof_mem_ranges_attr = 0 and vector_of_vma_mem_attr = NULL to retrieve
the number of memory ranges (vmas) and size of each memory range attribute.
Then, it allocates a buffer of that size and calls the ioctl again to fill
the buffer with memory range attributes.
This two-step interface allows userspace to first query the required
buffer size, then retrieve detailed attributes efficiently.
v2 (Matthew Brost)
- Use same ioctl to overload functionality
v3
- Add kernel-doc
v4
- Make uapi future proof by passing struct size (Matthew Brost)
- make lock interruptible (Matthew Brost)
- set reserved bits to zero (Matthew Brost)
- s/__copy_to_user/copy_to_user (Matthew Brost)
- Avod using VMA term in uapi (Thomas)
- xe_vm_put(vm) is missing (Shuicheng)
v5
- Nits
- Fix kernel-doc
Cc: Matthew Brost <matthew.brost@intel.com> Cc: Shuicheng Lin <shuicheng.lin@intel.com> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com> Reviewed-by: Matthew Brost <matthew.brost@intel.com> Link: https://lore.kernel.org/r/20250821173104.3030148-21-himal.prasad.ghimiray@intel.com Signed-off-by: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
drm/xe: Reset VMA attributes to default in SVM garbage collector
Restore default memory attributes for VMAs during garbage collection
if they were modified by madvise. Reuse existing VMA if fully overlapping;
otherwise, allocate a new mirror VMA.
v2 (Matthew Brost)
- Add helper for vma split
- Add retry to get updated vma
drm/xe/vm: Add helper to check for default VMA memory attributes
Introduce a new helper function `xe_vma_has_default_mem_attrs()` to
determine whether a VMA's memory attributes are set to their default
values. This includes checks for atomic access, PAT index, and preferred
location.
Also, add a new field `default_pat_index` to `struct xe_vma_mem_attr`
to track the initial PAT index set during the first bind. This helps
distinguish between default and user-modified pat index, such as those
changed via madvise.
drm/xe/madvise: Skip vma invalidation if mem attr are unchanged
If a VMA within the madvise input range already has the same memory
attribute as the one requested by the user, skip PTE zapping for that
VMA to avoid unnecessary invalidation.
v2 (Matthew Brost)
- fix skip_invalidation for new attributes
- s/u32/bool
- Remove unnecessary assignment for kzalloc'ed
A single BO can be linked to multiple VMAs, making VMA attributes
insufficient for determining the placement and PTE update attributes
of the BO. To address this, an attributes field has been added to the
BO.
drm/xe/svm: Add svm ranges migration policy on atomic access
If the platform does not support atomic access on system memory, and the
ranges are in system memory, but the user requires atomic accesses on
the VMA, then migrate the ranges to VRAM. Apply this policy for prefetch
operations as well.
v2
- Drop unnecessary vm_dbg
v3 (Matthew Brost)
- fix atomic policy
- prefetch shouldn't have any impact of atomic
- bo can be accessed from vma, avoid duplicate parameter
v4 (Matthew Brost)
- Remove TODO comment
- Fix comment
- Dont allow gpu atomic ops when user is setting atomic attr as CPU
This driver-specific ioctl enables UMDs to control the memory attributes
for GPU VMAs within a specified input range. If the start or end
addresses fall within an existing VMA, the VMA is split accordingly. The
attributes of the VMA are modified as provided by the users. The old
mappings of the VMAs are invalidated, and TLB invalidation is performed
if necessary.
v2(Matthew brost)
- xe_vm_in_fault_mode can't be enabled by Mesa, hence allow ioctl in non
fault mode too
- fix tlb invalidation skip for same ranges in multiple op
- use helper for tlb invalidation
- use xe_svm_notifier_lock/unlock helper
- s/lockdep_assert_held/lockdep_assert_held_write
- Add kernel-doc
v3(Matthew Brost)
- make vfunc fail safe
- Add sanitizing input args before vfunc
drm/xe: Allow CPU address mirror VMA unbind with gpu bindings for madvise
In the case of the MADVISE ioctl, if the start or end addresses fall
within a VMA and existing SVM ranges are present, remove the existing
SVM mappings. Then, continue with ops_parse to create new VMAs by REMAP
unmapping of old one.
v2 (Matthew Brost)
- Use vops flag to call unmapping of ranges in vm_bind_ioctl_ops_parse
- Rename the function
v3
- Fix doc
v4
- check if range is already in garbage collector (Matthew Brost)
drm/xe/vma: Modify new_vma to accept struct xe_vma_mem_attr as parameter
This change simplifies the logic by ensuring that remapped previous or
next VMAs are created with the same memory attributes as the original VMA.
By passing struct xe_vma_mem_attr as a parameter, we maintain consistency
in memory attributes.
-v2
*dst = *src (Matthew Brost)
-v3 (Matthew Brost)
Drop unnecessary helper
pass attr ptr as input to new_vma and vma_create
The attribute of xe_vma will determine the migration policy and the
encoding of the page table entries (PTEs) for that vma.
This attribute helps manage how memory pages are moved and how their
addresses are translated. It will be used by madvise to set the
behavior of the vma.
This commit introduces a new madvise interface to support
driver-specific ioctl operations. The madvise interface allows for more
efficient memory management by providing hints to the driver about the
expected memory usage and pte update policy for gpuvma.
v2 (Matthew/Thomas)
- Drop num_ops support
- Drop purgeable support
- Add kernel-docs
- IOWR/IOW
v3 (Matthew/Thomas)
- Reorder attributes
- use __u16 for migration_policy
- use __u64 for reserved in unions
- Avoid usage of vma
projects / thirdparty / kernel / stable.git / log
