- 16 Mar, 2022 10 commits
-
-
Matthew Auld authored
On integrated it looks like the GGTT base should always 1:1 maps to somewhere within DSM. On discrete the base seems to be pre-programmed with a normal lmem address, and is not 1:1 mapped with the base address. On such devices probe the lmem address directly from the PTE. v2(Ville): - The base is actually the pre-programmed GGTT address, which is then meant to 1:1 map to somewhere inside dsm. In the case of dgpu the base looks to just be some offset within lmem, but this also happens to be the exact dsm start, on dg1. Therefore we should only need to fudge the physical address, before allocating from stolen. - Bail if it's not located in dsm. v3: - Scratch that. There doesn't seem to be any relationship with the base and PTE address, on at least DG1. Let's instead just grab the lmem address from the PTE itself. Signed-off-by:Matthew Auld <matthew.auld@intel.com> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> Cc: Nirmoy Das <nirmoy.das@linux.intel.com> Reviewed-by:
Nirmoy Das <nirmoy.das@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20220315181425.576828-7-matthew.auld@intel.com
-
CQ Tang authored
When system does not have mappable aperture, ggtt->mappable_end=0. In this case if we pass PIN_MAPPABLE when pinning vma, the pinning code will return -ENOSPC. So conditionally set PIN_MAPPABLE if HAS_GMCH(). Suggested-by:
Chris P Wilson <chris.p.wilson@intel.com> Signed-off-by:
CQ Tang <cq.tang@intel.com> Cc: Radhakrishna Sripada <radhakrishna.sripada@intel.com> Cc: Ap Kamal <kamal.ap@intel.com> Reviewed-by:
-
Matthew Auld authored
For the ttm backend we can use existing placements fpfn and lpfn to force the allocator to place the object at the requested offset, potentially evicting stuff if the spot is currently occupied. Signed-off-by:
-
Matthew Auld authored
Add a generic interface for allocating an object at some specific offset, and convert stolen over. Later we will want to hook this up to different backends. Signed-off-by:
-
Matthew Auld authored
Keep the behaviour consistent with normal lmem, where we assume CPU access if by default required. Signed-off-by:
-
Akeem G Abodunrin authored
On client platforms with reduced LMEM BAR, we should be able to continue with driver load with reduced io_size. Instead of using the BAR size to determine the how large stolen should be, we should instead use the ADDR_RANGE register to figure this out(at least on platforms like DG2). For simplicity we don't attempt to support partially mappable stolen. v2: rearrange the io_mapping_init_wc slightly, since the stolen setup might result in reduced io_size. Signed-off-by:
Akeem G Abodunrin <akeem.g.abodunrin@intel.com> Co-developed-by:
-
Matthew Auld authored
Just pass along the probed io_size. The backend should be able to utilize the entire range here, even if some of it is non-mappable. It does leave open with what to do with stolen local-memory. Signed-off-by:
Thomas Hellström <thomas.hellstrom@linux.intel.com> Reviewed-by:
-
Matt Roper authored
Upcoming patches will need to steer writes to multicast registers as well as reading them. Although the setting of the 'multicast' bit should only really matter for write operations (reads always operate in a unicast manner and give us the result from one specific instance), Wa_22013088509 suggests that we leave the multicast bit enabled when performing read operations, so we follow suit here. Cc: Harish Chegondi <harish.chegondi@intel.com> Signed-off-by:
Matt Roper <matthew.d.roper@intel.com> Reviewed-by:
Lucas De Marchi <lucas.demarchi@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20220314234203.799268-4-matthew.d.roper@intel.com
-
Daniele Ceraolo Spurio authored
GuC has its own steering mechanism and can't use the default set by i915, so we need to provide the steering information that the FW will need to save/restore registers while processing an engine reset. The GUC interface allows us to do so as part of the register save/restore list and it requires us to specify the steering for all multicast register, even those that would be covered by the default setting for cpu access. Given that we do not distinguish between registers that do not need steering and registers that are guaranteed to work the default steering, we set the steering for all entries in the guc list that do not require a special steering (e.g. mslice) to the default settings; this will cost us a few extra writes during engine reset but allows us to keep the steering logic simple. Cc: John Harrison <John.C.Harrison@Intel.com> Cc: Matt Roper <matthew.d.roper@intel.com> Signed-off-by:
Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com> Signed-off-by:
-
Matt Roper authored
Add a new 'steering' node in each gt's debugfs directory that tells whether we're using explicit steering for various types of MCR ranges and, if so, what MMIO ranges it applies to. We're going to be transitioning away from implicit steering, even for slice/dss steering soon, so the information reported here will become increasingly valuable once that happens. v2: - Adding missing 'static' on intel_steering_types[] (Jose, sparse) v3: - "static const char *" -> "static const char * const" (sparse) Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com> Signed-off-by:
José Roberto de Souza <jose.souza@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20220315170250.954380-1-matthew.d.roper@intel.com
-
- 15 Mar, 2022 1 commit
-
-
John Harrison authored
sseu_dev_info is already a pretty large structure which will likely continue to grow when future platforms increase potential DSS and EU counts. Let's switch the stack placement of this structure in debugfs with a dynamic allocation. Signed-off-by:
John Harrison <John.C.Harrison@Intel.com> Signed-off-by:
-
- 14 Mar, 2022 2 commits
-
-
Matt Roper authored
When running on Xe_HP or beyond, let's use an updated format for describing topology in our error state dumps and debugfs to give a more accurate view of the hardware: - Just report DSS directly without the legacy "slice0" output that's no longer meaningful. - Indicate whether each DSS is accessible for geometry and/or compute. - Rename "rcs_topology" to "sseu_topology" since the information reported is common to both RCS and CCS engines now. v2: - Name static functions in a more consistent manner. (Lucas) Signed-off-by:
-
Matt Roper authored
Xe_HP removed "slice" as a first-class unit in the hardware design. Instead we now have a single pool of subslices (which are now referred to as "DSS") that different hardware units have different ways of grouping ("compute slices," "geometry slices," etc.). For the purposes of topology representation, we treat Xe_HP-based platforms as having a single slice that contains all of the platform's DSS. There's no need to allocate storage space for (max legacy slices * max dss); let's update some of our macros to minimize the storage requirement for sseu topology. We'll also document some of the constants to make it a little bit more clear what they represent. v2: - s/LEGACY/HSW/ in macro names. (Lucas) - Rename MAX() to SSEU_MAX() to avoid any potential clashes with other definitions elsewhere. Unfortunately max()/max_t() from linux/minmax.h cannot be used in this context. (Lucas) Signed-off-by:
-
- 11 Mar, 2022 2 commits
-
-
Matt Roper authored
We shouldn't really be keeping track of how many SFC_DONE registers our platforms can have, but rather how many SFC hardware units there can be (each SFC unit will have one corresponding SFC_DONE register). So drop the stray GEN12_SFC_DONE_MAX definition we had in the register definition file and replace it with an I915_MAX_SFC that follows the pattern we use for other hardware units. Note that our hardware has a 2:1:1 ratio of VD:VE:SFC, and as far as we know that pattern should carry forward to future platforms, so we'll define it as #VCS/2. Cc: Jani Nikula <jani.nikula@intel.com> Signed-off-by:
Jani Nikula <jani.nikula@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20220311062835.163744-1-matthew.d.roper@intel.com
-
Mastan Katragadda authored
A missing bounds check in vm_access() can lead to an out-of-bounds read or write in the adjacent memory area, since the len attribute is not validated before the memcpy later in the function, potentially hitting: [ 183.637831] BUG: unable to handle page fault for address: ffffc90000c86000 [ 183.637934] #PF: supervisor read access in kernel mode [ 183.637997] #PF: error_code(0x0000) - not-present page [ 183.638059] PGD 100000067 P4D 100000067 PUD 100258067 PMD 106341067 PTE 0 [ 183.638144] Oops: 0000 [#2] PREEMPT SMP NOPTI [ 183.638201] CPU: 3 PID: 1790 Comm: poc Tainted: G D 5.17.0-rc6-ci-drm-11296+ #1 [ 183.638298] Hardware name: Intel Corporation CoffeeLake Client Platform/CoffeeLake H DDR4 RVP, BIOS CNLSFWR1.R00.X208.B00.1905301319 05/30/2019 [ 183.638430] RIP: 0010:memcpy_erms+0x6/0x10 [ 183.640213] RSP: 0018:ffffc90001763d48 EFLAGS: 00010246 [ 183.641117] RAX: ffff888109c14000 RBX: ffff888111bece40 RCX: 0000000000000ffc [ 183.642029] RDX: 0000000000001000 RSI: ffffc90000c86000 RDI: ffff888109c14004 [ 183.642946] RBP: 0000000000000ffc R08: 800000000000016b R09: 0000000000000000 [ 183.643848] R10: ffffc90000c85000 R11: 0000000000000048 R12: 0000000000001000 [ 183.644742] R13: ffff888111bed190 R14: ffff888109c14000 R15: 0000000000001000 [ 183.645653] FS: 00007fe5ef807540(0000) GS:ffff88845b380000(0000) knlGS:0000000000000000 [ 183.646570] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 183.647481] CR2: ffffc90000c86000 CR3: 000000010ff02006 CR4: 00000000003706e0 [ 183.648384] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 [ 183.649271] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 [ 183.650142] Call Trace: [ 183.650988] <TASK> [ 183.651793] vm_access+0x1f0/0x2a0 [i915] [ 183.652726] __access_remote_vm+0x224/0x380 [ 183.653561] mem_rw.isra.0+0xf9/0x190 [ 183.654402] vfs_read+0x9d/0x1b0 [ 183.655238] ksys_read+0x63/0xe0 [ 183.656065] do_syscall_64+0x38/0xc0 [ 183.656882] entry_SYSCALL_64_after_hwframe+0x44/0xae [ 183.657663] RIP: 0033:0x7fe5ef725142 [ 183.659351] RSP: 002b:00007ffe1e81c7e8 EFLAGS: 00000246 ORIG_RAX: 0000000000000000 [ 183.660227] RAX: ffffffffffffffda RBX: 0000557055dfb780 RCX: 00007fe5ef725142 [ 183.661104] RDX: 0000000000001000 RSI: 00007ffe1e81d880 RDI: 0000000000000005 [ 183.661972] RBP: 00007ffe1e81e890 R08: 0000000000000030 R09: 0000000000000046 [ 183.662832] R10: 0000557055dfc2e0 R11: 0000000000000246 R12: 0000557055dfb1c0 [ 183.663691] R13: 00007ffe1e81e980 R14: 0000000000000000 R15: 0000000000000000 Changes since v1: - Updated if condition with range_overflows_t [Chris Wilson] Fixes: 9f909e21 ("drm/i915: Implement vm_ops->access for gdb access into mmaps") Signed-off-by:Mastan Katragadda <mastanx.katragadda@intel.com> Suggested-by:
Adam Zabrocki <adamza@microsoft.com> Reported-by:
Jackson Cody <cody.jackson@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Cc: Jon Bloomfield <jon.bloomfield@intel.com> Cc: Sudeep Dutt <sudeep.dutt@intel.com> Cc: <stable@vger.kernel.org> # v5.8+ Reviewed-by:
-
- 08 Mar, 2022 3 commits
-
-
Matt Roper authored
Platforms with FlatCCS do not use auxiliary planes for compression control data and thus do not need traditional aux table invalidation (and the registers no longer even exist). Original-author: CQ Tang Signed-off-by:
-
Matthew Auld authored
It looks like this code was accidentally dropped at some point(in a slightly different form), so add it back. The gist is that if we know the allocation will be one single chunk, then we can just annotate the BO with I915_BO_ALLOC_CONTIGUOUS, even if the user doesn't bother. In the future this should allow us to avoid using vmap for such objects, in some upcoming patches. v2(Thomas): - Tweak the commit message to mention the future motivation Signed-off-by:
-
Matthew Auld authored
Currently this will enforce both 2M alignment and padding for any LMEM pages inserted into the GGTT. However, this was only meant to be applied to the compact-pt layout with the ppGTT. For the GGTT we can reduce the alignment and padding to 64K. Bspec: 45015 Fixes: 87bd701e ("drm/i915: enforce min GTT alignment for discrete cards") Signed-off-by:
-
- 07 Mar, 2022 6 commits
-
-
Matthew Auld authored
This is no longer possible since e6e1a304 ("drm/i915: vma is always backed by an object."). Signed-off-by:
-
Matthew Auld authored
If the vm doesn't request async binding, like for example with the dpt, then we should be able to skip the async path and avoid calling i915_vm_lock_objects() altogether. Currently if we have a moving fence set for the BO(even though it might have signalled), we still take the async patch regardless of the bind_async setting, and then later still end up just doing i915_gem_object_wait_moving_fence() anyway. Alternatively we would need to add dummy scratch object which can be locked, just for the dpt. Suggested-by:
-
Matthew Auld authored
Since we are actually mapping the object and not the vma, when dealing with LMEM, we should be careful and use the backing store size here, since the vma->node.size could have all kinds of funny padding constraints, which could result in us writing to OOB address. v2(Chris): - Prefer vma->size here, which should be the backing store size. Some more rework is needed here to stop using node.size in some other places. Signed-off-by: -
Thomas Hellström authored
The test for vma should always return true, and when assigning -EBUSY to ret, the variable should already have that value. Signed-off-by:
Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20220304082641.308069-4-thomas.hellstrom@linux.intel.com
-
Thomas Hellström authored
Now that i915_vma_parked() is taking the object lock on vma destruction, and the only user of the vma refcount, i915_gem_object_unbind() also takes the object lock, remove the vma refcount. v3: Documentation update. Signed-off-by:
-
Thomas Hellström authored
vms are not getting properly closed. Rather than fixing that, Remove the vm open count and instead rely on the vm refcount. The vm open count existed solely to break the strong references the vmas had on the vms. Now instead make those references weak and ensure vmas are destroyed when the vm is destroyed. Unfortunately if the vm destructor and the object destructor both wants to destroy a vma, that may lead to a race in that the vm destructor just unbinds the vma and leaves the actual vma destruction to the object destructor. However in order for the object destructor to ensure the vma is unbound it needs to grab the vm mutex. In order to keep the vm mutex alive until the object destructor is done with it, somewhat hackishly grab a vm_resv refcount that is released late in the vma destruction process, when the vm mutex is no longer needed. v2: Address review-comments from Niranjana - Clarify that the struct i915_address_space::skip_pte_rewrite is a hack and should ideally be replaced in an upcoming patch. - Remove an unneeded continue in clear_vm_list and update comment. v3: - Documentation update - Commit message formatting Co-developed-by:
-
- 06 Mar, 2022 2 commits
-
-
Gwan-gyeong Mun authored
The current implementation of i915 prime mmap only works when initializing drm_i915_gem_object with shmem_region. When using LMEM, drm_i915_gem_object is initialized with ttm_system_region. In order to make prime mmap work even this case, when using LMEM (when using ttm in i915), dma_buf_ops.mmap callback function calls drm_gem_prime_mmap(). drm_gem_prime_mmap() of drm core calls internally i915_gem_mmap() so that prime mmap can perform normally. The fake offset is processed inside drm_gem_prime_mmap(). Testcase: igt/prime_mmap Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com> Cc: Matthew Auld <matthew.auld@intel.com> Signed-off-by:
Gwan-gyeong Mun <gwan-gyeong.mun@intel.com> Reviewed-by:
Nirmoy Das <nirmoy.das@intel.com> Signed-off-by:
Ramalingam C <ramalingam.c@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20220225131316.1433515-3-gwan-gyeong.mun@intel.com
-
Gwan-gyeong Mun authored
The dma_buf_ops.unmap_dma_buf callback used in i915, i915_gem_unmap_dma_buf(), has the same code as drm_gem_unmap_dma_buf(). In order to eliminate defining and using duplicate function, it updates the dma_buf_ops.unmap_dma_buf callback to use drm_gem_unmap_dma_buf(). Signed-off-by:
-
- 04 Mar, 2022 2 commits
-
-
Stuart Summers authored
If RCS is not enumerated, GuC will return invalid parameters. Make sure we do not send RCS supported when we have not enumerated it. Cc: Vinay Belgaumkar <vinay.belgaumkar@intel.com> Signed-off-by:
Stuart Summers <stuart.summers@intel.com> Signed-off-by:
-
Matt Roper authored
In the past we've always assumed that an RCS engine is present on every platform. However now that we have compute engines there may be platforms that have CCS engines but no RCS, or platforms that are designed to have both, but have the RCS engine fused off. Various engine-centric initialization that only needs to be done a single time for the group of RCS+CCS engines can't rely on being setup with the RCS now; instead we add a I915_ENGINE_FIRST_RENDER_COMPUTE flag that will be assigned to a single engine in the group; whichever engine has this flag will be responsible for some of the general setup (RCU_MODE programming, initialization of certain workarounds, etc.). Signed-off-by:
-
- 03 Mar, 2022 8 commits
-
-
John Harrison authored
Some G2H handlers were reading the context id field from the payload before checking the payload met the minimum length required. Signed-off-by:
-
John Harrison authored
The CTB registration process changed significantly a while back using a single KLV based H2G. So drop the original and now obsolete H2G definitions. Signed-off-by:
-
John Harrison authored
The LRC descriptor pool is going away. So, stop naming context ids as descriptor pool indecies. While at it, add a bunch of missing line feeds to some error messages. Signed-off-by:
-
John Harrison authored
The LRC descriptor was being initialised early on in the context registration sequence. It could then be determined that the actual registration needs to be delayed and the descriptor would be wiped out. This is inefficient, so move the setup to later in the process after the point of no return. v2: Move some split changes into the split patch (and do them correctly). Signed-off-by:
-
John Harrison authored
The LRC descriptor pool is going away. Further, the function that was populating it was also doing a bunch of logic about the context registration sequence. So, split that code apart into separate state setup and try to register functions. Note that some of those 'try to register' code paths actually undo the state setup and leave it to be redone again later (with potentially different values). This is inefficient. The next patch will correct this. Also, move a comment about ignoring return values to the place where the return values are actually ignored. v2: Move some more splitting from a later patch (and do it correctly). Signed-off-by:
-
John Harrison authored
The LRC descriptor pool is going away. So, stop using it as the limit for how many context ids are available. Instead, size the pool according to the number of contexts allowed. Note that this is just a naming change, the actual limit is identical in value. While at it, also update a kzalloc(sizeof()*count) to be a kcalloc(count,size). Signed-off-by:
-
John Harrison authored
The LRC descriptor pool is going away. So, stop using it as a check for whether submission has been initialised or not. Signed-off-by:
-
John Harrison authored
The LRC descriptor pool is going away. So, stop using it as a check for context registration, use the GuC id instead (being the thing that actually gets registered with the GuC). Also, rename the set/clear/query helper functions for context id mappings to better reflect their purpose and to differentiate from other registration related helper functions. Signed-off-by:
-
- 02 Mar, 2022 4 commits
-
-
Srinivasan Shanmugam authored
Registers that exist in the shared render/compute reset domain need to be placed on an engine workaround list to ensure that they are properly re-applied whenever an RCS or CCS engine is reset. We have a number of workarounds (updating registers MLTICTXCTL, L3SQCREG1_CCS0, GEN12_MERT_MOD_CTRL, and GEN12_GAMCNTRL_CTRL) that are incorrectly implemented on the 'gt' workaround list and need to be moved accordingly. Cc: Matt Roper <matthew.d.roper@intel.com> Signed-off-by:
Srinivasan Shanmugam <srinivasan.s@intel.com> Signed-off-by:
-
Matt Roper authored
Additional workarounds are required once we start exposing CCS engines. Note that we have a number of workarounds that update registers in the shared render/compute reset domain. Historically we've just added such registers to the RCS engine's workaround list. But going forward we should be more careful to place such workarounds on a wa_list for an engine that definitely exists and is not fused off (e.g., a platform with no RCS would never apply the RCS wa_list). We'll keep rcs_engine_wa_init() focused on RCS-specific workarounds that only need to be applied if the RCS engine is present. A separate general_render_compute_wa_init() function will be used to define workarounds that touch registers in the shared render/compute reset domain and that we need to apply regardless of what render and/or compute engines actually exist. Any workarounds defined in this new function will internally be added to the first present RCS or CCS engine's workaround list to ensure they get applied (and only get applied once rather than being needlessly re-applied several times). Co-author: Srinivasan Shanmugam Signed-off-by:
-
Daniele Ceraolo Spurio authored
HW resources are divided across the active CCS engines at the compute slice level, with each CCS having priority on one of the cslices. If a compute slice has no enabled DSS, its paired compute engine is not usable in full parallel execution because the other ones already fully saturate the HW, so consider it fused off. v2 (José): - moved it to its own function - fixed definition of ccs_mask v3 (Matt): - Replace fls() condition with a simple IP version test v4 (Matt): - Don't try to calculate a ccs_mask using intel_slicemask_from_dssmask() until we've determined that we're running on an Xe_HP platform where the logic makes sense (and won't overflow). Cc: Stuart Summers <stuart.summers@intel.com> Cc: Vinay Belgaumkar <vinay.belgaumkar@intel.com> Cc: Ashutosh Dixit <ashutosh.dixit@intel.com> Signed-off-by:
-
Matthew Brost authored
A different emit breadcrumbs ring programming is required for compute / render and we don't have UMD user so just reject parallel submission for these engine classes. Signed-off-by:
Matthew Brost <matthew.brost@intel.com> Signed-off-by:
-
