This patch would cause host deadlock when booting up a TDX VM even if huge page
is turned off. I currently reverted this patch. No further debug yet.
On Thu, Dec 12, 2024 at 12:36:33AM -0600, Michael Roth wrote:
> Currently the preparedness tracking relies on holding a folio's lock
> to keep allocations/preparations and corresponding updates to the
> prepared bitmap atomic.
>
> However, on the invalidation side, the bitmap entry for the GFN/index
> corresponding to a folio might need to be cleared after truncation. In
> these cases the folio's are no longer part of the filemap, so nothing
> guards against a newly-allocated folio getting prepared for the same
> GFN/index, and then subsequently having its bitmap entry cleared by the
> concurrently executing invalidation code.
>
> Avoid this by ensuring that the filemap invalidation lock is held to
> ensure allocations/preparations and corresponding updates to the
> prepared bitmap are atomic even versus invalidations. Use a shared lock
> in the kvm_gmem_get_pfn() case so vCPUs can still fault in pages in
> parallel.
>
> Signed-off-by: Michael Roth <michael.roth@xxxxxxx>
> ---
> virt/kvm/guest_memfd.c | 14 ++++++++++++++
> 1 file changed, 14 insertions(+)
>
> diff --git a/virt/kvm/guest_memfd.c b/virt/kvm/guest_memfd.c
> index 6907ae9fe149..9a5172de6a03 100644
> --- a/virt/kvm/guest_memfd.c
> +++ b/virt/kvm/guest_memfd.c
> @@ -154,6 +154,8 @@ static void kvm_gmem_mark_prepared(struct file *file, pgoff_t index, int order)
> unsigned long npages = (1ul << order);
> unsigned long *p;
>
> + rwsem_assert_held(&file->f_mapping->invalidate_lock);
> +
> /* The index isn't necessarily aligned to the requested order. */
> index &= ~(npages - 1);
> p = i_gmem->prepared + BIT_WORD(index);
> @@ -174,6 +176,8 @@ static void kvm_gmem_mark_range_unprepared(struct inode *inode, pgoff_t index, p
> struct kvm_gmem_inode *i_gmem = (struct kvm_gmem_inode *)inode->i_private;
> unsigned long *p = i_gmem->prepared + BIT_WORD(index);
>
> + rwsem_assert_held(&inode->i_mapping->invalidate_lock);
> +
> index &= BITS_PER_LONG - 1;
> if (index) {
> int first_word_count = min(npages, BITS_PER_LONG - index);
> @@ -200,6 +204,8 @@ static bool kvm_gmem_is_prepared(struct file *file, pgoff_t index, int order)
> unsigned long *p;
> bool ret;
>
> + rwsem_assert_held(&file->f_mapping->invalidate_lock);
> +
> /* The index isn't necessarily aligned to the requested order. */
> index &= ~(npages - 1);
> p = i_gmem->prepared + BIT_WORD(index);
> @@ -232,6 +238,8 @@ static int kvm_gmem_prepare_folio(struct kvm *kvm, struct file *file,
> pgoff_t index, aligned_index;
> int r;
>
> + rwsem_assert_held(&file->f_mapping->invalidate_lock);
> +
> index = gfn - slot->base_gfn + slot->gmem.pgoff;
> nr_pages = (1ull << max_order);
> WARN_ON(nr_pages > folio_nr_pages(folio));
> @@ -819,12 +827,16 @@ int kvm_gmem_get_pfn(struct kvm *kvm, struct kvm_memory_slot *slot,
> pgoff_t index = kvm_gmem_get_index(slot, gfn);
> struct file *file = kvm_gmem_get_file(slot);
> int max_order_local;
> + struct address_space *mapping;
> struct folio *folio;
> int r = 0;
>
> if (!file)
> return -EFAULT;
>
> + mapping = file->f_inode->i_mapping;
> + filemap_invalidate_lock_shared(mapping);
> +
> /*
> * The caller might pass a NULL 'max_order', but internally this
> * function needs to be aware of any order limitations set by
> @@ -838,6 +850,7 @@ int kvm_gmem_get_pfn(struct kvm *kvm, struct kvm_memory_slot *slot,
> folio = __kvm_gmem_get_pfn(file, slot, index, pfn, &max_order_local);
> if (IS_ERR(folio)) {
> r = PTR_ERR(folio);
> + filemap_invalidate_unlock_shared(mapping);
> goto out;
> }
>
> @@ -845,6 +858,7 @@ int kvm_gmem_get_pfn(struct kvm *kvm, struct kvm_memory_slot *slot,
> r = kvm_gmem_prepare_folio(kvm, file, slot, gfn, folio, max_order_local);
>
> folio_unlock(folio);
> + filemap_invalidate_unlock_shared(mapping);
>
> if (!r)
> *page = folio_file_page(folio, index);
> --
> 2.25.1
>
>
