- 21 Dec, 2023 40 commits
-
-
Daniele Ceraolo Spurio authored
GSC is only killed by an FLR, so we need to trigger one on unload to make sure we stop it. This is because we assign a chunk of memory to the GSC as part of the FW load, so we need to make sure it stops using it when we release it to the system on driver unload. Note that this is not a problem of the unload per-se, because the GSC will not touch that memory unless there are requests for it coming from the driver; therefore, no accesses will happen while Xe is not loaded, but if we re-load the driver then the GSC might wake up and try to access that old memory location again. Signed-off-by:
Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com> Cc: Alan Previn <alan.previn.teres.alexis@intel.com> Reviewed-by:
John Harrison <John.C.Harrison@Intel.com> Signed-off-by:
Rodrigo Vivi <rodrigo.vivi@intel.com>
-
Daniele Ceraolo Spurio authored
When the GSC FW is loaded, we need to inform it when a GSCCS reset is coming and then wait 200ms for it to get ready to process the reset. v2: move WA code to GSC file, use variable in Makefile (John) Signed-off-by:
-
Daniele Ceraolo Spurio authored
The GSC FW must be copied in a 4MB stolen memory allocation, whose GGTT address is then passed as a parameter to a dedicated load instruction submitted via the GSC engine. Since the GSC load is relatively slow (up to 250ms), we perform it asynchronously via a worker. This requires us to make sure that the worker has stopped before suspending/unloading. Note that we can't yet use xe_migrate_copy for the copy because it doesn't work with stolen memory right now, so we do a memcpy from the CPU side instead. v2: add comment about timeout value, fix GSC status checking before load (John) Bspec: 65306, 65346 Signed-off-by: -
Daniele Ceraolo Spurio authored
The GSC blob starts with a layout header, from which we can move to the boot directory, which in turns allows us to find the CPD. The CPD uses the same format as the one in the HuC binary, so we can re-use the same parsing code to get to the manifest, which contains the release and security versions of the FW. v2: Fix comments in struct definition (John) Signed-off-by:
-
Daniele Ceraolo Spurio authored
Add the basic definitions and init function. Same as HuC, GSC is only supported on the media GT on MTL and newer platforms. Note that the GSC requires submission resources which can't be allocated during init (because we don't have the hwconfig yet), so it can't be marked as loadable at the end of the init function. The allocation of those resources will come in the patch that makes use of them to load the FW. v2: better comment, move num FWs define inside the enum (John) Signed-off-by:
-
Daniele Ceraolo Spurio authored
The GSC firmware, support for which is coming soon for Xe, has both a release version (updated on every release) and a compatibility version (update only on interface changes). The GuC has something similar, with a global release version and a submission version (which is also known as the VF compatibility version). The main difference is that for the GuC we still want to check the driver requirement against the release version, while for the GSC we'll need to check against the compatibility version. Instead of special casing the GSC, this patch reworks the FW logic so that we store both versions at the uc_fw level for all binaries and we allow checking against either of the versions. Initially, we'll use it to support GSC, but the logic could be re-used to allow VFs to check against the GuC compatibility version. Note that the GSC version has 4 numbers (major, minor, hotfix, build), so support for that has been added as part of the rework and will be used in follow-up patches. Signed-off-by:
-
Bommithi Sakeena authored
Encapsulate all the module parameters in one single global struct variable. This also removes the extra xe_module.h from includes. v2: naming consistency as suggested by Jani and Lucas v3: fix checkpatch errors/warnings v4: adding blank line after struct declaration Cc: Jani Nikula <jani.nikula@linux.intel.com> Cc: Lucas De Marchi <lucas.demarchi@intel.com> Signed-off-by:
Bommithi Sakeena <bommithi.sakeena@intel.com> Reviewed-by:
Lucas De Marchi <lucas.demarchi@intel.com> Signed-off-by:
-
Tejas Upadhyay authored
This workaround applies to Xe2_LPG A0 V3: - Apply rule RENDER class V2(Matt): - Apply WA in lrc context Reviewed-by:
Matt Roper <matthew.d.roper@intel.com> Signed-off-by:
Tejas Upadhyay <tejas.upadhyay@intel.com> Signed-off-by:
-
Matt Roper authored
The workaround database has been updated to drop this workaround for all DG2 variants. Reviewed-by:
Gustavo Sousa <gustavo.sousa@intel.com> Link: https://lore.kernel.org/r/20231127190332.4099519-2-matthew.d.roper@intel.comSigned-off-by:
-
Lucas De Marchi authored
For Xe1 platforms, it's better to follow the way i915 adds the PCI IDs to the header, so it's easier to catch up when there is an update. This brings the same logic applied in commit 2e3c369f ("drm/i915/mtl: Eliminate subplatforms") to the equivalent xe header. The end result of this header for Xe1 platforms is now in sync with i915 as of commit 5032c607 ("drm/i915: ATS-M device ID update"). This can be seen by $ git show 5032c607:include/drm/i915_pciids.h > a.h $ git diff --color-words --no-index a.h include/drm/xe_pciids.h Reviewed-by:
-
Thomas Hellström authored
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.comSigned-off-by:
-
Thomas Hellström authored
xa_alloc_cyclic() returns 1 on successful allocation, if wrapping occurs, but the code incorrectly treats that as an error. Fix that. Also, xa_alloc_cyclic() requires xa_init_flags(..., XA_FLAGS_ALLOC), so fix that, and assuming we don't want a zero ASID, instead of using XA_FLAGS_ALLOC1, adjust the xa limits at alloc_cyclic time. v2: - On CONFIG_DRM_XE_DEBUG, Initialize the cyclic ASID allocation in such a way that the next allocated ASID will be the maximum one, and the one following will cause an ASID wrap, (all to have CI test high ASIDs and ASID wraps). v3: - Stricter return value checking from xa_alloc_cyclic() (Matthew Auld) Suggested-by:
Ohad Sharabi <osharabi@habana.ai> Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/946Signed-off-by:
Matthew Auld <matthew.auld@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20231124153345.97385-5-thomas.hellstrom@linux.intel.comSigned-off-by:
-
Thomas Hellström authored
trace_printk() is not intended for production code. Remove it. Suggested-by:
-
Thomas Hellström authored
Using "get" typically refers to obtaining a refcount, which we don't do here so rename to xe_bo_sg(). Suggested-by:
-
Thomas Hellström authored
Ensure, using xe_assert that the various try_add_<placement> functions don't access the bo placements array out-of-bounds. v2: - Remove the places argument to make sure the xe_assert operates on the array we're actually populating. (Matthew Auld) Suggested-by:
-
Michal Wajdeczko authored
We already print some basic information about the device, add virtualization information, until we expose that elsewhere. Reviewed-by:
Michal Wajdeczko <michal.wajdeczko@intel.com> Signed-off-by:
-
Michal Wajdeczko authored
We will be adding support for the SR-IOV and driver might be then running, in addition to existing non-virtualized bare-metal mode, also in Physical Function (PF) or Virtual Function (VF) mode. Since these additional modes require some changes to the driver, define enum flag to represent different SR-IOV modes and add a function where we will detect the actual mode in the runtime. We start with a forced bare-metal mode as it is sufficient to enable basic functionality and ensures no impact to existing code. Reviewed-by:
-
Michal Wajdeczko authored
The Single Root I/O Virtualization (SR-IOV) extension to the PCI Express (PCIe) specification suite is supported starting from 12th generation of Intel Graphics processors. Add a device flag that we will use to enable SR-IOV specific code paths and to indicate our readiness to support SR-IOV. We will enable this flag for the specific platforms once all required changes and additions will be ready and merged. Bspec: 52391 Reviewed-by:
-
Tejas Upadhyay authored
This workaround applies to Xe2_LPM V3(MattR): - Reorder reg and wa placement - Add base parameter to reg macro for better definition V2(MattR): - Change name of register - Loop for all engines - Driver permanent WA, applies to all steps Reviewed-by:
-
Tejas Upadhyay authored
This workaround applies to Xe2_LPM as well Reviewed-by:
-
Tejas Upadhyay authored
This workaround applies to Xe2_LPM Reviewed-by:
-
Gustavo Sousa authored
With the current implementation, a preemption or other kind of interrupt might happen between xe_mmio_read32() and ktime_get_raw(). Such an interruption (specially in the case of preemption) might be long enough to cause a timeout without giving a chance of a new check on the register value on a next iteration, which would have happened otherwise. This issue causes some sporadic timeouts in some code paths. As an example, we were experiencing some rare timeouts when waiting for PLL unlock for C10/C20 PHYs (see intel_cx0pll_disable()). After debugging, we found out that the PLL unlock was happening within the expected time period (20us), which suggested a bug in xe_mmio_wait32(). To fix the issue, ensure that we do a last check out of the loop if necessary. This change was tested with the aforementioned PLL unlocking code path. Experiments showed that, before this change, we observed reported timeouts in 54 of 5000 runs; and, after this change, no timeouts were reported in 5000 runs. v2: - Prefer an implementation without a barrier (v1 switched the order of xe_mmio_read32() and ktime_get_raw() calls and added a barrier() in between). (Lucas, Rodrigo) Cc: Rodrigo Vivi <rodrigo.vivi@intel.com> Reviewed-by: -
Gustavo Sousa authored
This function is big enough, let's move it to a shared compilation unit. While at it, document it. Here is the output of running bloat-o-metter on the new and old module (execution provided by Lucas): $ ./scripts/bloat-o-meter build64/drivers/gpu/drm/xe/xe.ko{.old,} add/remove: 2/0 grow/shrink: 0/58 up/down: 554/-15645 (-15091) (...) # Lines in between omitted Total: Before=2181322, After=2166231, chg -0.69% The overall reduction in the size is not that significant. Nevertheless, keeping the function as inline arguably does not bring too much benefit as well. As noted by Lucas, we would probably benefit from an inline function that did the fast-path check: do an optimistic first check before entering the wait-logic, which itself would go to a compilation unit. We might come back to implement this in the future if we have data to justify it. v2: - Add note in documentation for @timeout_us regarding the exponential backoff strategy. (Lucas) - Share output of bloat-o-meter in the commit message. (Lucas) Reviewed-by: -
Ruthuvikas Ravikumar authored
This kunit verifies the hardware values of mocs and l3cc registers with the KMD programmed values. v14: Fix CHECK. v13: Remove ret after forcewake. v11: Add KUNIT_ASSERT_EQ_MSG for Forcewake. v9/v10: Add Forcewake Fail. v8: Remove xe_bo.h and xe_pm.h Remove mocs and l3cc from live_mocs. Pull debug and err msg for mocs/l3cc out of if else block. Add HAS_LNCF_MOCS. v7: correct checkpath v6: Change ssize_t type. Change forcewake domain to XE_FW_GT. Update change of MOCS registers are multicast on Xe_HP and beyond patch. v5: Release forcewake. Remove single statement braces. Fix debug statements. v4: Drop stratch and vaddr. Fix debug statements. Fix indentation. v3: Fix checkpath. v2: Fix checkpath. Cc: Aravind Iddamsetty <aravind.iddamsetty@intel.com> Cc: Mathew D Roper <matthew.d.roper@intel.com> Reviewed-by:Ruthuvikas Ravikumar <ruthuvikas.ravikumar@intel.com> Link: https://lore.kernel.org/r/20231116215152.2248859-1-ruthuvikas.ravikumar@intel.comSigned-off-by:
-
Lucas De Marchi authored
When built with W=1, the following warnings show up on modpost: MODPOST drivers/gpu/drm/xe/Module.symvers WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/gpu/drm/xe/tests/xe_bo_test.o WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/gpu/drm/xe/tests/xe_dma_buf_test.o WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/gpu/drm/xe/tests/xe_migrate_test.o WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/gpu/drm/xe/tests/xe_pci_test.o WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/gpu/drm/xe/tests/xe_rtp_test.o WARNING: modpost: missing MODULE_DESCRIPTION() in drivers/gpu/drm/xe/tests/xe_wa_test.o Add the module description for each of these to fix the warning. Reviewed-by:
-
Haridhar Kalvala authored
ATS-M device ID update. BSpec: 44477 Signed-off-by:
Haridhar Kalvala <haridhar.kalvala@intel.com> Reviewed-by:
-
José Roberto de Souza authored
Those are ids present in i915 but missing in Xe. Cc: Lucas De Marchi <lucas.demarchi@intel.com> Reviewed-by:
José Roberto de Souza <jose.souza@intel.com> Signed-off-by:
-
José Roberto de Souza authored
RPL-U is defined as a subplatform but those PCI ids were not included in pciidlist so Xe KMD would never probe device with those ids. This is following what i915 does to include RPL-U to PCI ids probe list. v2: - change order to match i915 Cc: Lucas De Marchi <lucas.demarchi@intel.com> Reviewed-by:
-
Matthew Brost authored
DRM_XE_VM_BIND_OP_MAP_* IOCTL operations can result in GPUVA unmap, remap, or map operations in vm_bind_ioctl_ops_create. The xe_vma_op.map fields are blindly set which is incorrect for GPUVA unmap or remap operations. Fix this by only setting xe_vma_op.map for GPUVA map operations. Also restructure a bit vm_bind_ioctl_ops_create to make the code a bit more readable. Reported-by:
Dafna Hirschfeld <dhirschfeld@habana.ai> Signed-off-by:
Matthew Brost <matthew.brost@intel.com> Reviewed-by:
Brian Welty <brian.welty@intel.com> Signed-off-by:
-
Lucas De Marchi authored
After noticing in logs there were still mentions to GEN6 registers, it was clear commit d9b79ad2 ("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:
-
Lucas De Marchi authored
The mocs documentation was copied from i915 and doesn't match the reality in xe. Reword it so it matches what the code is doing. Reviewed-by:
-
Lucas De Marchi authored
GEN11_MOCS_ENTRIES dates back from importing the table from the i915 module. The macro was used so the it could be maintained in a single place and platforms would just override with additional entries. With the platforms supported by xe, each of them is just defining individual tables without re-using this define. Move it inside gen12_mocs_desc that is the only user. Reviewed-by:
-
Matthew Auld authored
This seems to create a locking inversion with object_name_lock. The lock is held by drm_prime_fd_to_handle when calling our xe_gem_prime_import hook, which might eventually go on to grab the dma-resv lock during the attach. However we also have the opposite locking order in xe_gem_create_ioctl which is holding the dma-resv lock when calling drm_gem_handle_create, which wants to eventually grab object_name_lock: -> #1 (reservation_ww_class_mutex){+.+.}-{3:3}: <4> [635.739288] lock_acquire+0x169/0x3d0 <4> [635.739294] __ww_mutex_lock.constprop.0+0x164/0x1e60 <4> [635.739300] ww_mutex_lock_interruptible+0x42/0x1a0 <4> [635.739305] drm_gem_shmem_pin+0x4b/0x140 [drm_shmem_helper] <4> [635.739317] dma_buf_dynamic_attach+0x101/0x430 <4> [635.739323] xe_gem_prime_import+0xcc/0x2e0 [xe] <4> [635.739499] drm_prime_fd_to_handle_ioctl+0x184/0x2e0 [drm] <4> [635.739594] drm_ioctl_kernel+0x16f/0x250 [drm] <4> [635.739693] drm_ioctl+0x35e/0x620 [drm] <4> [635.739789] __x64_sys_ioctl+0xb7/0xf0 <4> [635.739794] do_syscall_64+0x3c/0x90 <4> [635.739799] entry_SYSCALL_64_after_hwframe+0x6e/0xd8 <4> [635.739805] -> #0 (&dev->object_name_lock){+.+.}-{3:3}: <4> [635.739813] check_prev_add+0x1ba/0x14a0 <4> [635.739818] __lock_acquire+0x203e/0x2ff0 <4> [635.739823] lock_acquire+0x169/0x3d0 <4> [635.739827] __mutex_lock+0x124/0x1310 <4> [635.739832] drm_gem_handle_create+0x32/0x50 [drm] <4> [635.739927] xe_gem_create_ioctl+0x1d3/0x550 [xe] <4> [635.740102] drm_ioctl_kernel+0x16f/0x250 [drm] <4> [635.740197] drm_ioctl+0x35e/0x620 [drm] <4> [635.740293] __x64_sys_ioctl+0xb7/0xf0 <4> [635.740297] do_syscall_64+0x3c/0x90 <4> [635.740302] entry_SYSCALL_64_after_hwframe+0x6e/0xd8 <4> [635.740307] It looks like it should be safe to simply drop the dma-resv lock prior to publishing the object when calling drm_gem_handle_create. Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/743Signed-off-by: -
Michal Wajdeczko authored
There are only 4 scratch registers VF_SW_FLAG(0..3) on each GuC. We shouldn't use non-existing register VF_SW_FLAG(4) for posting read. Signed-off-by:
-
Michal Wajdeczko authored
If GuC responds with the NO_RESPONSE_BUSY message, we extend our timeout while waiting for the actual response, but we wrongly assumed that the next message will be RESPONSE_SUCCESS, missing that we still can get RESPONSE_FAILURE. Change the condition for the expected message type, using only common bits from RESPONSE_SUCCESS and RESPONSE_FAILURE (as they differ, by ABI design, only by the last bit). v2: add comment/checks to the code (Matt) Signed-off-by:
-
Michal Wajdeczko authored
While copying GuC response from the scratch registers to the buffer, formula to identify next scratch register is broken. Fix it. Signed-off-by:
-
Michal Wajdeczko authored
This variable holds full length of the message, including header length so it should be checked against GUC_CTB_MSG_MAX_LEN. Signed-off-by:
-
Matt Roper authored
The workaround database was just updated to extend this workaround to DG2-G11 (whereas previously it applied only to G10 and G12). Reviewed-by:
-
Rodrigo Vivi authored
Let's bring a bit of clarity on this 'region' field that is part of vm_bind operation struct. Rename and document to make it more than obvious that it is a region instance and not a mask and also that it should only be used with the prefetch operation itself. Signed-off-by:
Francois Dugast <francois.dugast@intel.com> Reviewed-by:
-
Rodrigo Vivi authored
On one hand the WAIT_OP represents the operation use for waiting such as ==, !=, > and so on. On the other hand, the mask is applied to the value used for comparision. Split those two to bring clarity to the uapi. Signed-off-by:
-
