diff options
| author | Arthur Hanson Rasmusson <arthur@vgpu.io> | 2026-10-05 12:47:47 -0400 |
|---|---|---|
| committer | Arthur Hanson Rasmusson <arthur@vgpu.io> | 2026-10-05 12:47:47 -0400 |
| commit | 0f227d063e281c4b8ec21e61132ae1ddf0920053 (patch) | |
| tree | aba72956d7ab9d6cb69212696f56091af77281ab /kernel/bar_emulation.c | |
| parent | 708a3802824358c1770d5ad0b3c2e24fae114eb8 (diff) | |
| download | open-iov-0f227d063e281c4b8ec21e61132ae1ddf0920053.tar.gz open-iov-0f227d063e281c4b8ec21e61132ae1ddf0920053.zip | |
vgsp: fix per-type VRAM type lookup and canonical FB-size encoding
The 708a3802 per-type VRAM change stripped an nvidia- prefix from
mdev->type->sysfs_name, but sysfs_name is the registered type name
verbatim; the mdev core only prefixes the type DIRECTORY with the parent
driver string. Every lookup failed, so all mdevs silently fell back to
512 MiB, nvidia-vgpu-turing-1g was unreachable and available_instances
capped at 1. Match sysfs_name directly in get_available,
show_description and mdev_create.
Also canonicalise the NV_PFB_PRI_MMU_LOCAL_MEMORY_RANGE encoder
(BAR0+0x100ce0) by stripping factors of two from MAG so 512 MiB encodes
0x19 again (1 GiB 0x1a, 2 GiB 0x1b) instead of the non-canonical 0x204.
Default-512M behaviour is now byte-equivalent to the known-good pre-VRAM
module except max_instances=2. Module sha256 cb1724d3.
Diffstat (limited to 'kernel/bar_emulation.c')
| -rw-r--r-- | kernel/bar_emulation.c | 17 |
1 files changed, 14 insertions, 3 deletions
diff --git a/kernel/bar_emulation.c b/kernel/bar_emulation.c index 8e3f8cf..d1702c8 100644 --- a/kernel/bar_emulation.c +++ b/kernel/bar_emulation.c @@ -290,13 +290,24 @@ int nv_vgpu_bar_emul_init(struct nv_vgpu_device *vgpu) * guest's fbSize - reserved arithmetic stays consistent. */ u64 fb = nv_vgpu_fb_emul_size(ctx->vgpu); - u32 scale = 0; + u32 scale = 0, mag; + /* + * Find the largest SCALE whose MAG fits the 6-bit field, then + * canonicalise: strip factors of two from MAG while SCALE can + * absorb them. 512 MiB -> mag=1,scale=9 -> 0x19 (the historical + * known-good value); 1 GiB -> 0x1a; 2 GiB -> 0x1b. Sizes that + * are not exactly representable round down. + */ while (scale < 15 && (fb >> (scale + 20)) > 0x3f) scale++; + mag = (u32)(fb >> (scale + 20)); + while (scale < 15 && mag > 1 && !(mag & 1u)) { + mag >>= 1; + scale++; + } *(u32 *)((u8 *)ctx->bar0_shadow + 0x100ce0) = - ((((u32)(fb >> (scale + 20))) & 0x3f) << 4) | - (scale & 0xf); + ((mag & 0x3f) << 4) | (scale & 0xf); } /* |