From: Dave Airlie <[email protected]> When NVK enabled large pages userspace tests were seeing fault reports at a valid address.
There was a case where an address moving from 64k page to 4k pages could expose a race between unmapping the 4k page, mapping the 64k page and unref the 4k pages. Unref 4k pages would cause the dual-page table handling to always set the LPTE entry to SPARSE or INVALID, but if we'd mapped a valid LPTE in the meantime, it would get trashed. Keep track of when a valid LPTE has been referenced, and don't reset in that case. This increase the tracking to 32-bit, because it turns out if unref can get delayed, you can get a lot of these outstanding and this can cause strange behaviours. Cc: [email protected] Link: https://gitlab.freedesktop.org/mesa/mesa/-/issues/14610 Signed-off-by: Dave Airlie <[email protected]> -- v2: move to 32-bit from 8-bit tracker fix some more flag changes. --- drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.c | 43 +++++++++++++------ drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.h | 9 ++-- 2 files changed, 36 insertions(+), 16 deletions(-) diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.c index f95c58b67633..304aaed9767d 100644 --- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.c +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.c @@ -53,7 +53,7 @@ nvkm_vmm_pt_new(const struct nvkm_vmm_desc *desc, bool sparse, } } - if (!(pgt = kzalloc(sizeof(*pgt) + lpte, GFP_KERNEL))) + if (!(pgt = kzalloc(sizeof(*pgt) + (sizeof(pgt->pte[0]) * lpte), GFP_KERNEL))) return NULL; pgt->page = page ? page->shift : 0; pgt->sparse = sparse; @@ -242,14 +242,17 @@ nvkm_vmm_unref_sptes(struct nvkm_vmm_iter *it, struct nvkm_vmm_pt *pgt, if (pgt->pte[pteb] & NVKM_VMM_PTE_SPARSE) { TRA(it, "LPTE %05x: U -> S %d PTEs", pteb, ptes); pair->func->sparse(vmm, pgt->pt[0], pteb, ptes); - } else - if (pair->func->invalid) { - /* If the MMU supports it, restore the LPTE to the - * INVALID state to tell the MMU there is no point - * trying to fetch the corresponding SPTEs. - */ - TRA(it, "LPTE %05x: U -> I %d PTEs", pteb, ptes); - pair->func->invalid(vmm, pgt->pt[0], pteb, ptes); + } else if (!(pgt->pte[pteb] & NVKM_VMM_PTE_BIG_VALID)) { + if (pair->func->invalid) { + /* If the MMU supports it, restore the LPTE to the + * INVALID state to tell the MMU there is no point + * trying to fetch the corresponding SPTEs. + */ + TRA(it, "LPTE %05x: U -> I %d PTEs", pteb, ptes); + pair->func->invalid(vmm, pgt->pt[0], pteb, ptes); + } + } else { + TRA(it, "LPTE %05x: V %d PTEs", pteb, ptes); } } } @@ -280,6 +283,13 @@ nvkm_vmm_unref_ptes(struct nvkm_vmm_iter *it, bool pfn, u32 ptei, u32 ptes) if (desc->type == SPT && (pgt->refs[0] || pgt->refs[1])) nvkm_vmm_unref_sptes(it, pgt, desc, ptei, ptes); + if (desc->type == LPT && (pgt->refs[0] || pgt->refs[1])) { + for (u32 lpti = ptei; ptes; lpti++) { + pgt->pte[lpti] &= ~NVKM_VMM_PTE_BIG_VALID; + ptes--; + } + } + /* PT no longer needed? Destroy it. */ if (!pgt->refs[type]) { it->lvl++; @@ -336,6 +346,7 @@ nvkm_vmm_ref_sptes(struct nvkm_vmm_iter *it, struct nvkm_vmm_pt *pgt, if (pgt->pte[ptei] & NVKM_VMM_PTE_VALID) break; pgt->pte[ptei] |= NVKM_VMM_PTE_VALID; + pgt->pte[ptei] &= ~NVKM_VMM_PTE_BIG_VALID; } if (pgt->pte[pteb] & NVKM_VMM_PTE_SPARSE) { @@ -374,6 +385,14 @@ nvkm_vmm_ref_ptes(struct nvkm_vmm_iter *it, bool pfn, u32 ptei, u32 ptes) if (desc->type == SPT) nvkm_vmm_ref_sptes(it, pgt, desc, ptei, ptes); + if (desc->type == LPT) { + for (u32 lpti = ptei; ptes; lpti++) { + pgt->pte[lpti] &= ~NVKM_VMM_PTE_VALID; + pgt->pte[lpti] |= NVKM_VMM_PTE_BIG_VALID; + ptes--; + } + } + return true; } @@ -386,7 +405,7 @@ nvkm_vmm_sparse_ptes(const struct nvkm_vmm_desc *desc, pgt->pde[ptei++] = NVKM_VMM_PDE_SPARSE; } else if (desc->type == LPT) { - memset(&pgt->pte[ptei], NVKM_VMM_PTE_SPARSE, ptes); + memset32(&pgt->pte[ptei], NVKM_VMM_PTE_SPARSE, ptes); } } @@ -398,7 +417,7 @@ nvkm_vmm_sparse_unref_ptes(struct nvkm_vmm_iter *it, bool pfn, u32 ptei, u32 pte memset(&pt->pde[ptei], 0x00, sizeof(pt->pde[0]) * ptes); else if (it->desc->type == LPT) - memset(&pt->pte[ptei], 0x00, sizeof(pt->pte[0]) * ptes); + memset32(&pt->pte[ptei], 0x00, ptes); return nvkm_vmm_unref_ptes(it, pfn, ptei, ptes); } @@ -457,7 +476,7 @@ nvkm_vmm_ref_hwpt(struct nvkm_vmm_iter *it, struct nvkm_vmm_pt *pgd, u32 pdei) desc->func->sparse(vmm, pt, pteb, ptes); else desc->func->invalid(vmm, pt, pteb, ptes); - memset(&pgt->pte[pteb], 0x00, ptes); + memset32(&pgt->pte[pteb], 0x00, ptes); } else { desc->func->unmap(vmm, pt, pteb, ptes); while (ptes--) diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.h b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.h index 4586a425dbe4..3720579f4bf7 100644 --- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.h +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.h @@ -44,10 +44,11 @@ struct nvkm_vmm_pt { * * This information is used to manage LPTE state transitions. */ -#define NVKM_VMM_PTE_SPARSE 0x80 -#define NVKM_VMM_PTE_VALID 0x40 -#define NVKM_VMM_PTE_SPTES 0x3f - u8 pte[]; +#define NVKM_VMM_PTE_SPARSE 0x80000000 +#define NVKM_VMM_PTE_VALID 0x40000000 +#define NVKM_VMM_PTE_BIG_VALID 0x20000000 +#define NVKM_VMM_PTE_SPTES 0x1fffffff + u32 pte[]; }; typedef void (*nvkm_vmm_pxe_func)(struct nvkm_vmm *, -- 2.52.0
