[5/8] drm/vmwgfx: Support huge page faults

Commit Message

Thomas Hellström (Intel) Dec. 3, 2019, 1:22 p.m. UTC

From: Thomas Hellstrom <[e-mail]>

With vmwgfx dirty-tracking we need a specialized huge_fault
callback. Implement and hook it up.

Cc: Andrew Morton <[e-mail]>
Cc: Michal Hocko <[e-mail]>
Cc: "Matthew Wilcox (Oracle)" <[e-mail]>
Cc: "Kirill A. Shutemov" <[e-mail]>
Cc: Ralph Campbell <[e-mail]>
Cc: "Jérôme Glisse" <[e-mail]>
Cc: "Christian König" <[e-mail]>
Signed-off-by: Thomas Hellstrom <[e-mail]>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h        |  2 +
 drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c | 66 +++++++++++++++++++++-
 drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c   |  1 +
 3 files changed, 68 insertions(+), 1 deletion(-)

Patch

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index a31e726d6d71..8656a97448c3 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -1428,6 +1428,8 @@  void vmw_bo_dirty_unmap(struct vmw_buffer_object *vbo,
 			pgoff_t start, pgoff_t end);
 vm_fault_t vmw_bo_vm_fault(struct vm_fault *vmf);
 vm_fault_t vmw_bo_vm_mkwrite(struct vm_fault *vmf);
+vm_fault_t vmw_bo_vm_huge_fault(struct vm_fault *vmf,
+				enum page_entry_size pe_size);
 
 /**
  * VMW_DEBUG_KMS - Debug output for kernel mode-setting
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c b/drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c
index 17a5dca7b921..6f76a97ad969 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_page_dirty.c
@@ -473,7 +473,7 @@  vm_fault_t vmw_bo_vm_fault(struct vm_fault *vmf)
 	 * a lot of unnecessary write faults.
 	 */
 	if (vbo->dirty && vbo->dirty->method == VMW_BO_DIRTY_MKWRITE)
-		prot = vma->vm_page_prot;
+		prot = vm_get_page_prot(vma->vm_flags & ~VM_SHARED);
 	else
 		prot = vm_get_page_prot(vma->vm_flags);
 
@@ -486,3 +486,67 @@  vm_fault_t vmw_bo_vm_fault(struct vm_fault *vmf)
 
 	return ret;
 }
+
+vm_fault_t vmw_bo_vm_huge_fault(struct vm_fault *vmf,
+				enum page_entry_size pe_size)
+{
+	struct vm_area_struct *vma = vmf->vma;
+	struct ttm_buffer_object *bo = (struct ttm_buffer_object *)
+	    vma->vm_private_data;
+	struct vmw_buffer_object *vbo =
+		container_of(bo, struct vmw_buffer_object, base);
+	pgprot_t prot;
+	vm_fault_t ret;
+	pgoff_t fault_page_size;
+
+	switch (pe_size) {
+	case PE_SIZE_PMD:
+		fault_page_size = HPAGE_PMD_SIZE >> PAGE_SHIFT;
+		break;
+	case PE_SIZE_PUD:
+		fault_page_size = HPAGE_PUD_SIZE >> PAGE_SHIFT;
+		break;
+	default:
+		WARN_ON_ONCE(1);
+		return VM_FAULT_FALLBACK;
+	}
+
+	/* Always do write dirty-tracking on PTE level. */
+	if (READ_ONCE(vbo->dirty) && (vmf->flags & FAULT_FLAG_WRITE))
+		return VM_FAULT_FALLBACK;
+
+	ret = ttm_bo_vm_reserve(bo, vmf);
+	if (ret)
+		return ret;
+
+	if (vbo->dirty) {
+		pgoff_t allowed_prefault;
+		unsigned long page_offset;
+
+		page_offset = vmf->pgoff -
+			drm_vma_node_start(&bo->base.vma_node);
+		if (page_offset >= bo->num_pages ||
+		    vmw_resources_clean(vbo, page_offset,
+					page_offset + PAGE_SIZE,
+					&allowed_prefault)) {
+			ret = VM_FAULT_SIGBUS;
+			goto out_unlock;
+		}
+
+		/*
+		 * Write protect, so we get a new fault on write, and can
+		 * split.
+		 */
+		prot = vm_get_page_prot(vma->vm_flags & ~VM_SHARED);
+	} else {
+		prot = vm_get_page_prot(vma->vm_flags);
+	}
+	ret = ttm_bo_vm_fault_reserved(vmf, prot, 1, fault_page_size);
+	if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT))
+		return ret;
+
+out_unlock:
+	dma_resv_unlock(bo->base.resv);
+
+	return ret;
+}
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c b/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c
index ce288756531b..de838ba88a97 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_glue.c
@@ -33,6 +33,7 @@  int vmw_mmap(struct file *filp, struct vm_area_struct *vma)
 		.pfn_mkwrite = vmw_bo_vm_mkwrite,
 		.page_mkwrite = vmw_bo_vm_mkwrite,
 		.fault = vmw_bo_vm_fault,
+		.huge_fault = vmw_bo_vm_huge_fault,
 		.open = ttm_bo_vm_open,
 		.close = ttm_bo_vm_close
 	};