diff options
| author | Arthur Hanson Rasmusson <arthur@vgpu.io> | 2026-09-22 05:16:37 -0400 |
|---|---|---|
| committer | Arthur Hanson Rasmusson <arthur@vgpu.io> | 2026-09-22 05:16:37 -0400 |
| commit | 9b891254b05e9e46f27b8ff53decfbbb429d1cc4 (patch) | |
| tree | 0b2584b1e533c0706fc538b47067bd66194e9c03 /kernel/main.c | |
| parent | 915f5e5c8c86c0846816b4d0e1c2f633bf1c51d5 (diff) | |
| download | open-iov-9b891254b05e9e46f27b8ff53decfbbb429d1cc4.tar.gz open-iov-9b891254b05e9e46f27b8ff53decfbbb429d1cc4.zip | |
kernel: dedicated host-RAM vidmem buffer for BAR1 (not host GPU's real VRAM)
Guest VRAM (BAR1) was mapped to the host GPU's real BAR1 (managed by
nvidia.ko) - guest VRAM accesses hit the broken host GPU. Now a
dedicated 256MB vzalloc buffer backs guest VRAM; BAR1 mmap uses
per-page vmalloc_to_pfn (like BAR0 shadow). vfree on cleanup.
Diffstat (limited to 'kernel/main.c')
| -rw-r--r-- | kernel/main.c | 24 |
1 files changed, 18 insertions, 6 deletions
diff --git a/kernel/main.c b/kernel/main.c index a47fe23..5e99881 100644 --- a/kernel/main.c +++ b/kernel/main.c @@ -1141,18 +1141,30 @@ static int nv_vgpu_vfio_dev_mmap(struct vfio_device *vdev, struct vm_area_struct return 0; } - /* BAR1 region - VRAM (use physical address directly) */ + /* BAR1 region - VRAM (emulated vidmem buffer, per-page vmalloc_to_pfn) */ region_base = VFIO_PCI_INDEX_TO_OFFSET(VFIO_PCI_BAR1_REGION_INDEX); region_size = vgpu->bars[1].size; if (offset >= region_base && offset < region_base + region_size) { - if (region_size == 0) + unsigned long vram_off = offset - region_base; + unsigned long pos; + + if (!vgpu->bars[1].vaddr || region_size == 0) return -ENODEV; if (offset + size > region_base + region_size) return -EINVAL; - /* Use physical address directly since vaddr is NULL (managed by nvidia.ko) */ - pfn = (vgpu->bars[1].gpa + (offset - region_base)) >> PAGE_SHIFT; - vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); - return remap_pfn_range(vma, vma->vm_start, pfn, size, vma->vm_page_prot); + /* vidmem buffer is vzalloc'd: physically non-contiguous pages, so + * map one page at a time via vmalloc_to_pfn() (like BAR0 shadow). + * Guest VRAM accesses hit our clean host-RAM backing, not the + * host GPU's real BAR1. */ + vm_flags_set(vma, VM_DONTEXPAND | VM_DONTDUMP); + vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot); + for (pos = 0; pos < size; pos += PAGE_SIZE) { + pfn = vmalloc_to_pfn(vgpu->bars[1].vaddr + vram_off + pos); + if (remap_pfn_range(vma, vma->vm_start + pos, pfn, PAGE_SIZE, + vma->vm_page_prot)) + return -EAGAIN; + } + return 0; } /* BAR2 region - ioremap'd host BAR; map backing physical address */ |
