Re: [PATCH 2/2] mm/hugetlb: add wrapper functions for interactions with page cache

[Mike Kravetz <mike.kravetz@oracle.com>]

On 06/09/23 12:52, Sidhartha Kumar wrote:
> On 6/9/23 12:49 PM, Sidhartha Kumar wrote:
> > Signed-off-by: Sidhartha Kumar <sidhartha.kumar@oracle.com>
> 
> Sorry, I missed adding the commit message to this. It should be:
> 
> Add filemap_lock_hugetlb_folio() which is wraps __filemap_get_folio()
> and passes in a linear page index. hugetlb_add_to_page_cache() is modified
> to also compute a linear page index before calling into page cache code.
> 
> linear_page_index() is modified to perform the computation on hugetlb
> so we can use it in the page cache wrappers.
> 
> > ---
> >   fs/hugetlbfs/inode.c    | 14 +++++++-------
> >   include/linux/hugetlb.h | 21 +++++++++++++++++++--
> >   include/linux/pagemap.h |  2 --
> >   mm/hugetlb.c            | 22 +++++++++++++---------
> >   4 files changed, 39 insertions(+), 20 deletions(-)
> > 
> > diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
> > index 90361a922cec7..90d27a8af4b6a 100644
> > --- a/fs/hugetlbfs/inode.c
> > +++ b/fs/hugetlbfs/inode.c
> > @@ -617,20 +617,19 @@ static void remove_inode_hugepages(struct inode *inode, loff_t lstart,
> >   	struct hstate *h = hstate_inode(inode);
> >   	struct address_space *mapping = &inode->i_data;
> >   	const pgoff_t start = lstart >> huge_page_shift(h);
> > -	const pgoff_t end = lend >> huge_page_shift(h);
> >   	struct folio_batch fbatch;
> >   	pgoff_t next, index;
> >   	int i, freed = 0;
> >   	bool truncate_op = (lend == LLONG_MAX);
> >   	folio_batch_init(&fbatch);
> > -	next = start;
> > -	while (filemap_get_folios(mapping, &next, end - 1, &fbatch)) {
> > +	next = lstart;
> > +	while (filemap_get_folios(mapping, &next, lend - 1, &fbatch)) {

This does not seem correct.  At this point next == lstart which is a file
offset passed to the routine as opposed to an index.

I would think next needs to be set to 'lstart >> PAGE_SHIFT' here.

> >   		for (i = 0; i < folio_batch_count(&fbatch); ++i) {
> >   			struct folio *folio = fbatch.folios[i];
> >   			u32 hash = 0;
> > -			index = folio->index;
> > +			index = (folio->index) >> huge_page_shift(h);

Here you want to convert index from the PAGE_SIZE index to a hugetlb
page size index.  Correct?
I am terrible at arithmetic, but huge_page_shift already includes
PAGE_SHIFT, so it seems like you want this to me.

			index = (folio->index) >> huge_page_order(h);

> >   			hash = hugetlb_fault_mutex_hash(mapping, index);
> >   			mutex_lock(&hugetlb_fault_mutex_table[hash]);
> > @@ -693,10 +692,11 @@ static void hugetlbfs_zero_partial_page(struct hstate *h,
> >   					loff_t start,
> >   					loff_t end)
> >   {
> > -	pgoff_t idx = start >> huge_page_shift(h);
> > +	struct mm_struct *mm = current->mm;
> > +	struct vm_area_struct *vma = find_vma(mm, start);
> >   	struct folio *folio;
> > -	folio = filemap_lock_folio(mapping, idx);
> > +	folio = filemap_lock_hugetlb_folio(vma, start);

Here you are passing an address/index that may be associated with a
tail page.  I assume since the hugetlb folio is multi-order, the
returned folio will be for the 'head page'.  Correct?

> >   	if (IS_ERR(folio))
> >   		return;
> > @@ -868,7 +868,7 @@ static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset,
> >   		}
> >   		clear_huge_page(&folio->page, addr, pages_per_huge_page(h));
> >   		__folio_mark_uptodate(folio);
> > -		error = hugetlb_add_to_page_cache(folio, mapping, index);
> > +		error = hugetlb_add_to_page_cache(folio, &pseudo_vma, mapping, addr);
> >   		if (unlikely(error)) {
> >   			restore_reserve_on_error(h, &pseudo_vma, addr, folio);
> >   			folio_put(folio);
> > diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
> > index 21f942025fecd..55f90e051b7a2 100644
> > --- a/include/linux/hugetlb.h
> > +++ b/include/linux/hugetlb.h
> > @@ -727,8 +727,8 @@ struct folio *alloc_hugetlb_folio_nodemask(struct hstate *h, int preferred_nid,
> >   				nodemask_t *nmask, gfp_t gfp_mask);
> >   struct folio *alloc_hugetlb_folio_vma(struct hstate *h, struct vm_area_struct *vma,
> >   				unsigned long address);
> > -int hugetlb_add_to_page_cache(struct folio *folio, struct address_space *mapping,
> > -			pgoff_t idx);
> > +int hugetlb_add_to_page_cache(struct folio *folio, struct vm_area_struct *vma,
> > +				struct address_space *mapping, unsigned long address);
> >   void restore_reserve_on_error(struct hstate *h, struct vm_area_struct *vma,
> >   				unsigned long address, struct folio *folio);
> > @@ -755,6 +755,16 @@ static inline struct hugepage_subpool *hugetlb_folio_subpool(struct folio *folio
> >   	return folio->_hugetlb_subpool;
> >   }
> > +/* Wrapper function for __filemap_get_folio*/
> > +static inline struct folio *filemap_lock_hugetlb_folio(struct vm_area_struct *vma,
> > +						unsigned long address)
> > +{
> > +	struct address_space *mapping = vma->vm_file->f_mapping;
> > +
> > +	pgoff_t idx = linear_page_index(vma, address);
> > +	return __filemap_get_folio(mapping, idx, FGP_LOCK, 0);
> > +}

I like the wrapper idea.  This is going to replace existing calls to
filemap_lock_folio.  What about something like this for the routine:

static inline struct folio *filemap_lock_hugetlb_folio(struct hstate *h,
					struct address_space *mapping,
					pgoff_t index);
{
	/* please verify my arithmetic */
	return filemap_lock_folio(mapping, index << huge_page_order(h));
}

In this way, existing callers would only need to be changed to pass in
hstate.  Perhaps, there was a reason for your wrapper not obvious to me?

> > +
> >   static inline void hugetlb_set_folio_subpool(struct folio *folio,
> >   					struct hugepage_subpool *subpool)
> >   {
> > @@ -1021,6 +1031,13 @@ static inline struct hugepage_subpool *hugetlb_folio_subpool(struct folio *folio
> >   	return NULL;
> >   }
> > +/* Wrapper function for __filemap_get_folio*/
> > +static inline struct folio *filemap_lock_hugetlb_folio(struct vm_area_struct *vma,
> > +						unsigned long address)
> > +{
> > +	return NULL;
> > +}
> > +
> >   static inline int isolate_or_dissolve_huge_page(struct page *page,
> >   						struct list_head *list)
> >   {
> > diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
> > index 17c414fc2136e..ae8f36966d7b3 100644
> > --- a/include/linux/pagemap.h
> > +++ b/include/linux/pagemap.h
> > @@ -860,8 +860,6 @@ static inline pgoff_t linear_page_index(struct vm_area_struct *vma,
> >   					unsigned long address)
> >   {
> >   	pgoff_t pgoff;
> > -	if (unlikely(is_vm_hugetlb_page(vma)))
> > -		return linear_hugepage_index(vma, address);
> >   	pgoff = (address - vma->vm_start) >> PAGE_SHIFT;
> >   	pgoff += vma->vm_pgoff;
> >   	return pgoff;
> > diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> > index dfa412d8cb300..824d6d215a161 100644
> > --- a/mm/hugetlb.c
> > +++ b/mm/hugetlb.c
> > @@ -951,7 +951,7 @@ static long region_count(struct resv_map *resv, long f, long t)
> >   /*
> >    * Convert the address within this vma to the page offset within
> > - * the mapping, in pagecache page units; huge pages here.
> > + * the mapping, in huge page units here.
> >    */
> >   static pgoff_t vma_hugecache_offset(struct hstate *h,
> >   			struct vm_area_struct *vma, unsigned long address)
> > @@ -5730,7 +5730,7 @@ static bool hugetlbfs_pagecache_present(struct hstate *h,
> >   			struct vm_area_struct *vma, unsigned long address)
> >   {
> >   	struct address_space *mapping = vma->vm_file->f_mapping;
> > -	pgoff_t idx = vma_hugecache_offset(h, vma, address);
> > +	pgoff_t idx = linear_page_index(vma, address);
> >   	bool present;
> >   	rcu_read_lock();
> > @@ -5740,13 +5740,16 @@ static bool hugetlbfs_pagecache_present(struct hstate *h,
> >   	return present;
> >   }
> > -int hugetlb_add_to_page_cache(struct folio *folio, struct address_space *mapping,
> > -			   pgoff_t idx)
> > +int hugetlb_add_to_page_cache(struct folio *folio,
> > +			struct vm_area_struct *vma,
> > +			struct address_space *mapping,
> > +			unsigned long address)
> >   {

Like filemap_lock_hugetlb_folio(), can you just add hstate to existing
hugetlb_add_to_page_cache() arguments and do arithmetic to convert index
to PAGE_SIZE based index?

Again, I could be missing something.  But, IMO such a conversion would
minimize changes to the current code.
-- 
Mike Kravetz

> >   	struct inode *inode = mapping->host;
> >   	struct hstate *h = hstate_inode(inode);
> >   	int err;
> > +	pgoff_t idx = linear_page_index(vma, address);
> >   	__folio_set_locked(folio);
> >   	err = __filemap_add_folio(mapping, folio, idx, GFP_KERNEL, NULL);
> > @@ -5854,7 +5857,8 @@ static vm_fault_t hugetlb_no_page(struct mm_struct *mm,
> >   	 * before we get page_table_lock.
> >   	 */
> >   	new_folio = false;
> > -	folio = filemap_lock_folio(mapping, idx);
> > +
> > +	folio = filemap_lock_hugetlb_folio(vma, address);
> >   	if (IS_ERR(folio)) {
> >   		size = i_size_read(mapping->host) >> huge_page_shift(h);
> >   		if (idx >= size)
> > @@ -5913,7 +5917,7 @@ static vm_fault_t hugetlb_no_page(struct mm_struct *mm,
> >   		new_folio = true;
> >   		if (vma->vm_flags & VM_MAYSHARE) {
> > -			int err = hugetlb_add_to_page_cache(folio, mapping, idx);
> > +			int err = hugetlb_add_to_page_cache(folio, vma, mapping, address);
> >   			if (err) {
> >   				/*
> >   				 * err can't be -EEXIST which implies someone
> > @@ -6145,7 +6149,7 @@ vm_fault_t hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
> >   		/* Just decrements count, does not deallocate */
> >   		vma_end_reservation(h, vma, haddr);
> > -		pagecache_folio = filemap_lock_folio(mapping, idx);
> > +		pagecache_folio = filemap_lock_hugetlb_folio(vma, address);
> >   		if (IS_ERR(pagecache_folio))
> >   			pagecache_folio = NULL;
> >   	}
> > @@ -6258,7 +6262,7 @@ int hugetlb_mfill_atomic_pte(pte_t *dst_pte,
> >   	if (is_continue) {
> >   		ret = -EFAULT;
> > -		folio = filemap_lock_folio(mapping, idx);
> > +		folio = filemap_lock_hugetlb_folio(dst_vma, dst_addr);
> >   		if (IS_ERR(folio))
> >   			goto out;
> >   		folio_in_pagecache = true;
> > @@ -6350,7 +6354,7 @@ int hugetlb_mfill_atomic_pte(pte_t *dst_pte,
> >   		 * hugetlb_fault_mutex_table that here must be hold by
> >   		 * the caller.
> >   		 */
> > -		ret = hugetlb_add_to_page_cache(folio, mapping, idx);
> > +		ret = hugetlb_add_to_page_cache(folio, dst_vma, mapping, dst_addr);
> >   		if (ret)
> >   			goto out_release_nounlock;
> >   		folio_in_pagecache = true;
>