On 2022.12.19 20:52:04 +0800, Zheng Wang wrote:
> If intel_gvt_dma_map_guest_page failed, it will call
> ppgtt_invalidate_spt, which will finally free the spt. But the caller does
> not notice that, it will free spt again in error path.
>
It's not clear from this description which caller is actually wrong,
better to clarify the problem in ppgtt_populate_spt_by_guest_entry() function.
> Fix this by undoing the mapping of DMA address and freeing sub_spt.
>
> Fixes: b901b252b6cf ("drm/i915/gvt: Add 2M huge gtt support")
> Signed-off-by: Zheng Wang <zyytlz.wz@xxxxxxx>
> ---
> v4:
> - fix by undo the mapping of DMA address and free sub_spt suggested by Zhi
>
> v3:
> - correct spelling mistake and remove unused variable suggested by Greg
>
> v2: https://lore.kernel.org/all/20221006165845.1735393-1-zyytlz.wz@xxxxxxx/
>
> v1: https://lore.kernel.org/all/20220928033340.1063949-1-zyytlz.wz@xxxxxxx/
> ---
> drivers/gpu/drm/i915/gvt/gtt.c | 53 +++++++++++++++++++++++++++++-----
> 1 file changed, 46 insertions(+), 7 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/gvt/gtt.c b/drivers/gpu/drm/i915/gvt/gtt.c
> index 51e5e8fb505b..b472e021e5a4 100644
> --- a/drivers/gpu/drm/i915/gvt/gtt.c
> +++ b/drivers/gpu/drm/i915/gvt/gtt.c
> @@ -1192,11 +1192,11 @@ static int split_2MB_gtt_entry(struct intel_vgpu *vgpu,
> {
> const struct intel_gvt_gtt_pte_ops *ops = vgpu->gvt->gtt.pte_ops;
> struct intel_vgpu_ppgtt_spt *sub_spt;
> - struct intel_gvt_gtt_entry sub_se;
> + struct intel_gvt_gtt_entry sub_se, e;
> unsigned long start_gfn;
> dma_addr_t dma_addr;
> - unsigned long sub_index;
> - int ret;
> + unsigned long sub_index, parent_index;
> + int ret, ret1;
>
> gvt_dbg_mm("Split 2M gtt entry, index %lu\n", index);
>
> @@ -1209,10 +1209,8 @@ static int split_2MB_gtt_entry(struct intel_vgpu *vgpu,
> for_each_shadow_entry(sub_spt, &sub_se, sub_index) {
> ret = intel_gvt_dma_map_guest_page(vgpu, start_gfn + sub_index,
> PAGE_SIZE, &dma_addr);
> - if (ret) {
> - ppgtt_invalidate_spt(spt);
> - return ret;
> - }
> + if (ret)
> + goto err;
I think it's fine to remove this and leave to upper caller, but again please
describe the behavior change in commit message as well, e.g to fix the sanity
of spt destroy that leaving previous invalidate and free of spt to caller function
instead of within callee function.
> sub_se.val64 = se->val64;
>
> /* Copy the PAT field from PDE. */
> @@ -1231,6 +1229,47 @@ static int split_2MB_gtt_entry(struct intel_vgpu *vgpu,
> ops->set_pfn(se, sub_spt->shadow_page.mfn);
> ppgtt_set_shadow_entry(spt, se, index);
> return 0;
> +err:
> + /* Undone the existing mappings of DMA addr. */
> + for_each_present_shadow_entry(spt, &e, parent_index) {
sub_spt? We're undoing what's mapped for sub_spt right?
> + switch (e.type) {
> + case GTT_TYPE_PPGTT_PTE_4K_ENTRY:
> + gvt_vdbg_mm("invalidate 4K entry\n");
> + ppgtt_invalidate_pte(spt, &e);
> + break;
> + case GTT_TYPE_PPGTT_PTE_64K_ENTRY:
> + /* We don't setup 64K shadow entry so far. */
> + WARN(1, "suspicious 64K gtt entry\n");
> + continue;
> + case GTT_TYPE_PPGTT_PTE_2M_ENTRY:
> + gvt_vdbg_mm("invalidate 2M entry\n");
> + continue;
> + case GTT_TYPE_PPGTT_PTE_1G_ENTRY:
> + WARN(1, "GVT doesn't support 1GB page\n");
> + continue;
> + case GTT_TYPE_PPGTT_PML4_ENTRY:
> + case GTT_TYPE_PPGTT_PDP_ENTRY:
> + case GTT_TYPE_PPGTT_PDE_ENTRY:
I don't think this all entry type makes sense, as here we just split
2M entry for multiple 4K PTE entry.
> + gvt_vdbg_mm("invalidate PMUL4/PDP/PDE entry\n");
> + ret1 = ppgtt_invalidate_spt_by_shadow_entry(
> + spt->vgpu, &e);
> + if (ret1) {
> + gvt_vgpu_err("fail: shadow page %p shadow entry 0x%llx type %d\n",
> + spt, e.val64, e.type);
> + goto free_spt;
> + }
for above reason, I don't think this is valid.
> + break;
> + default:
> + GEM_BUG_ON(1);
> + }
> + }
> + /* Release the new alloced apt. */
> +free_spt:
> + trace_spt_change(sub_spt->vgpu->id, "release", sub_spt,
> + sub_spt->guest_page.gfn, sub_spt->shadow_page.type);
> + ppgtt_free_spt(sub_spt);
> + sub_spt = NULL;
> + return ret;
> }
>
> static int split_64KB_gtt_entry(struct intel_vgpu *vgpu,
> --
> 2.25.1
>
Attachment:
signature.asc
Description: PGP signature
