On Wed, Nov 04, 2026 at 11:39:30AM +0000, Matthew Auld wrote: > On 03/11/2026 20:21, Matthew Brost wrote: > > On Tue, Nov 03, 2026 at 07:00:24PM +0000, Matthew Auld wrote: > > > On 11/07/2026 03:56, Matthew Brost wrote: > > > > Add support for mapping a ttm_tt's pages using the IOVA-based DMA API > > > > (dma_iova_try_alloc()/link()/sync()/unlink()/free()) as an alternative > > > > to the scatter-gather list. An IOVA mapping is a single contiguous DMA > > > > range, which is cheaper to walk and program than an sg list. > > > > > > > > Whether the IOVA API can be used depends on the IOMMU mode, so the > > > > decision is made once per tt at xe_ttm_tt_create() time and recorded in > > > > struct xe_ttm_tt::use_iova. To keep the scope limited, only device BOs > > > > that are not GGTT-mapped are eligible: dma-buf imports manage their own > > > > mapping and GGTT bindings are programmed directly from the sg list. > > > > dma_iova_try_alloc() may still fail, in which case the tt transparently > > > > falls back to the sg list, which remains the default. > > > > > > > > The IOVA reservation is allocated at tt creation and freed at tt > > > > destroy. xe_tt_map_iova() links the tt pages (coalescing physically > > > > contiguous runs) and syncs the IOTLB; xe_tt_unmap_iova() unlinks them > > > > but keeps the reservation so the tt can be remapped without > > > > re-allocating IOVA space. xe_tt_map()/xe_tt_unmap() dispatch between the > > > > IOVA and sg backends, and xe_bo_dma_unmap_pinned() learns about the IOVA > > > > case. > > > > > > > > Consumers that walked the sg list via xe_res_first_sg(xe_bo_sg()) now go > > > > through the new xe_res_first_tt() helper, which initializes either an > > > > IOVA cursor (xe_res_first_iova()) or an sg cursor depending on the tt's > > > > mapping mode. > > > > > > > > Cc: Carlos Santa <[email protected]> > > > > Cc: Ryan Neph <[email protected]> > > > > Cc: Christian Koenig <[email protected]> > > > > Cc: Huang Rui <[email protected]> > > > > Cc: Matthew Auld <[email protected]> > > > > Cc: Maarten Lankhorst <[email protected]> > > > > Cc: Maxime Ripard <[email protected]> > > > > Cc: Thomas Zimmermann <[email protected]> > > > > Cc: David Airlie <[email protected]> > > > > Cc: Simona Vetter <[email protected]> > > > > Cc: [email protected] > > > > Cc: [email protected] > > > > Cc: Thomas Hellström <[email protected]> > > > > Assisted-by: GitHub_Copilot:claude-opus-4.8 > > > > Signed-off-by: Matthew Brost <[email protected]> > > > > > > > > > <snip> > > > > > > > static void xe_ttm_tt_destroy(struct ttm_device *ttm_dev, struct > > > > ttm_tt *tt) > > > > { > > > > + struct xe_ttm_tt *xe_tt = container_of(tt, struct xe_ttm_tt, > > > > ttm); > > > > + > > > > + /* > > > > + * Release the IOVA reservation made in xe_ttm_tt_create(). The > > > > pages > > > > + * must already have been unlinked by xe_tt_unmap(). > > > > + */ > > > > + if (xe_tt->use_iova) { > > > > + struct xe_device *xe = ttm_to_xe_device(ttm_dev); > > > > + > > > > + xe_assert(xe, !xe_tt->iova_linked); > > > > + dma_iova_free(xe->drm.dev, &xe_tt->iova); > > > > > > Do we know if this is allowed? I think there is xe_bo_pci_dev_remove_all() > > > which requires that evict_all will correctly tear down iommu mappings on > > > stuff like pci unplug and I don't think tt_destroy is called in that flow? > > > > But unpopulate is called which unlinks / syncs (TLB flush for the IOVA). > > > > > Does having the iova here cause issues? Also, for VRAM flow I think there > > > is > > > a tt_create() via pipeline gutting which will now create a new iova? > > > > > > > Maybe we shuffle this around a bit so the iova alloc is tied to > > populate/unpopulate. It was been a while I forget exactly why I did this > > way. > > > > > Aside from that, maybe we can land 27/28 as standalone? > > > > Yes, let me extract these two into standalone patches today and > > reworking if needed. > > Quickly double checked this, indeed looks to be needed: >
Thanks was able recreate this + verify a local fix. 27/28 has one TTM dependency [1] and also is staged behind dma mapping accounting [2]. So maybe once [2] lands an updated 27/28 can get posted as standalone. Matt [1] https://patchwork.freedesktop.org/patch/739339/?series=170216&rev=2 [2] https://patchwork.freedesktop.org/series/177790/ > [ 808.058892] [IGT] core_hotunplug: finished subtest > hotreplug-lateclose-with-load, SUCCESS > [ 808.058928] [IGT] core_hotunplug: exiting, ret=0 > [ 808.059673] BUG: kernel NULL pointer dereference, address: > [ 808.061068] RIP: 0010:iommu_get_dma_domain+0x10/0x20 > [ 808.061141] PKRU: 55555554 > [ 808.061144] Call Trace: > [ 808.061146] <TASK> > [ 808.061149] dma_iova_free+0x18/0xf0 > [ 808.061153] xe_ttm_tt_destroy+0x23/0x40 [xe] > [ 808.061231] ttm_bo_put+0x242/0x3d0 [ttm] > [ 808.061412] ? drm_gpuvm_bo_put+0x199/0x200 [drm_gpuvm] > [ 808.061418] xe_vma_destroy_late+0x89/0x90 [xe] > [ 808.061476] xe_vm_close_and_put+0x5de/0x920 [xe] > [ 808.061534] xe_file_close+0x105/0x190 [xe] > [ 808.061712] drm_file_free+0x218/0x260 > [ 808.061717] drm_release_noglobal+0x6d/0xd0 > [ 808.061721] __fput+0xf9/0x270 > [ 808.061725] task_work_run+0x8b/0xb0 > [ 808.061730] do_exit+0x229/0x8b0 > [ 808.061735] do_group_exit+0x8b/0x90 > [ 808.061739] __x64_sys_exit_group+0x17/0x20 > [ 808.061744] x64_sys_call+0x18c6/0x18d0 > [ 808.061749] do_syscall_64+0xee/0x350 > [ 808.061753] ? __pte_offset_map+0x1d/0xd0 > [ 808.061757] ? next_uptodate_folio+0xbb/0x3a0 > [ 808.061762] ? set_pte_range+0x20d/0x290 > [ 808.061766] ? filemap_map_pages+0x8e1/0x9c0 > [ 808.061772] ? count_memcg_events+0x7c/0x1b0 > [ 808.061776] ? handle_mm_fault+0xbde/0x14f0 > [ 808.061781] ? __se_sys_close+0x7a/0xe0 > [ 808.061785] ? __se_sys_close+0x7a/0xe0 > [ 808.061790] ? do_user_addr_fault+0x272/0x6c0 > [ 808.061794] ? irqentry_exit+0x3f/0x640 > [ 808.061799] ? do_syscall_64+0x42/0x350 > [ 808.061803] entry_SYSCALL_64_after_hwframe+0x76/0x7e > [ 808.061808] RIP: 0033:0x7f95428c71b8 > > I assume this is dev->iommu_group = NULL post unplug. > > > > > Matt >
