Re: [RFC PATCH v1 09/37] KVM: guest_memfd: Skip LRU for guest_memfd folios

[Yan Zhao <yan.y.zhao@intel.com>]

> +static struct folio *__kvm_gmem_get_folio(struct address_space *mapping,
> +					  pgoff_t index,
> +					  struct mempolicy *policy)
> +{
> +	const gfp_t gfp = mapping_gfp_mask(mapping);
> +	struct folio *folio;
> +	int err;
> +
> +	folio = filemap_lock_folio(mapping, index);
> +	if (!IS_ERR(folio))
> +		return folio;
> +
> +	folio = filemap_alloc_folio(gfp, 0, policy);
> +	if (!folio)
> +		return ERR_PTR(-ENOMEM);
> +
> +	err = mem_cgroup_charge(folio, NULL, gfp);
> +	if (err)
> +		goto err_put;
> +
> +	__folio_set_locked(folio);
> +
> +	err = __filemap_add_folio(mapping, folio, index, gfp, NULL);
> +	if (err) {
		 mem_cgroup_uncharge(folio);

> +		__folio_clear_locked(folio);
> +		goto err_put;
> +	}
> +
> +	return folio;
> +
> +err_put:
> +	folio_put(folio);
> +	return ERR_PTR(err);
> +}
> +
>  /*
>   * Returns a locked folio on success.  The caller is responsible for
>   * setting the up-to-date flag before the memory is mapped into the guest.
> @@ -160,6 +195,7 @@ static struct mempolicy *kvm_gmem_get_folio_policy(struct gmem_inode *gi,
>  static struct folio *kvm_gmem_get_folio(struct inode *inode, pgoff_t index)
>  {
>  	/* TODO: Support huge pages. */
> +	struct address_space *mapping = inode->i_mapping;
>  	struct mempolicy *policy;
>  	struct folio *folio;
>  
> @@ -167,16 +203,17 @@ static struct folio *kvm_gmem_get_folio(struct inode *inode, pgoff_t index)
>  	 * Fast-path: See if folio is already present in mapping to avoid
>  	 * policy_lookup.
>  	 */
> -	folio = filemap_lock_folio(inode->i_mapping, index);
> +	folio = filemap_lock_folio(mapping, index);
>  	if (!IS_ERR(folio))
>  		return folio;
>  
>  	policy = kvm_gmem_get_folio_policy(GMEM_I(inode), index);
> -	folio = __filemap_get_folio_mpol(inode->i_mapping, index,
> -					 FGP_LOCK | FGP_CREAT,
> -					 mapping_gfp_mask(inode->i_mapping), policy);
> -	mpol_cond_put(policy);
>  
> +	do {
> +		folio = __kvm_gmem_get_folio(mapping, index, policy);
> +	} while (IS_ERR(folio) && PTR_ERR(folio) == -EEXIST);
Why not just return ERR_PTR(-EEXIST) up to kvm_gmem_get_pfn() and have a higher
level retry?

> +	mpol_cond_put(policy);
>  	return folio;
>  }