Re: [PATCH v2] mm, thp: Relax the VM_DENYWRITE constraint on file-backed THPs

[Hugh Dickins]

On Mon, 5 Apr 2021, Collin Fijalkovich wrote:

> Transparent huge pages are supported for read-only non-shmem files,
> but are only used for vmas with VM_DENYWRITE. This condition ensures that
> file THPs are protected from writes while an application is running
> (ETXTBSY).  Any existing file THPs are then dropped from the page cache
> when a file is opened for write in do_dentry_open(). Since sys_mmap
> ignores MAP_DENYWRITE, this constrains the use of file THPs to vmas
> produced by execve().
> 
> Systems that make heavy use of shared libraries (e.g. Android) are unable
> to apply VM_DENYWRITE through the dynamic linker, preventing them from
> benefiting from the resultant reduced contention on the TLB.
> 
> This patch reduces the constraint on file THPs allowing use with any
> executable mapping from a file not opened for write (see
> inode_is_open_for_write()). It also introduces additional conditions to
> ensure that files opened for write will never be backed by file THPs.
> 
> Restricting the use of THPs to executable mappings eliminates the risk that
> a read-only file later opened for write would encounter significant
> latencies due to page cache truncation.
> 
> The ld linker flag '-z max-page-size=(hugepage size)' can be used to
> produce executables with the necessary layout. The dynamic linker must
> map these file's segments at a hugepage size aligned vma for the mapping to
> be backed with THPs.
> 
> Comparison of the performance characteristics of 4KB and 2MB-backed
> libraries follows; the Android dex2oat tool was used to AOT compile an
> example application on a single ARM core.
> 
> 4KB Pages:
> ==========
> 
> count              event_name            # count / runtime
> 598,995,035,942    cpu-cycles            # 1.800861 GHz
>  81,195,620,851    raw-stall-frontend    # 244.112 M/sec
> 347,754,466,597    iTLB-loads            # 1.046 G/sec
>   2,970,248,900    iTLB-load-misses      # 0.854122% miss rate
> 
> Total test time: 332.854998 seconds.
> 
> 2MB Pages:
> ==========
> 
> count              event_name            # count / runtime
> 592,872,663,047    cpu-cycles            # 1.800358 GHz
>  76,485,624,143    raw-stall-frontend    # 232.261 M/sec
> 350,478,413,710    iTLB-loads            # 1.064 G/sec
>     803,233,322    iTLB-load-misses      # 0.229182% miss rate
> 
> Total test time: 329.826087 seconds
> 
> A check of /proc/$(pidof dex2oat64)/smaps shows THPs in use:
> 
> /apex/com.android.art/lib64/libart.so
> FilePmdMapped:      4096 kB
> 
> /apex/com.android.art/lib64/libart-compiler.so
> FilePmdMapped:      2048 kB
> 
> Signed-off-by: Collin Fijalkovich <cfijalkovich@xxxxxxxxxx>

Acked-by: Hugh Dickins <hughd@xxxxxxxxxx>

and you also won

Reviewed-by: William Kucharski <william.kucharski@xxxxxxxxxx>

in the v1 thread.

I had hoped to see a more dramatic difference in the numbers above,
but I'm a performance naif, and presume other loads and other
libraries may show further benefit.

> ---
> Changes v1 -> v2:
> * commit message 'non-shmem filesystems' -> 'non-shmem files'
> * Add performance testing data to commit message
> 
>  fs/open.c       | 13 +++++++++++--
>  mm/khugepaged.c | 16 +++++++++++++++-
>  2 files changed, 26 insertions(+), 3 deletions(-)
> 
> diff --git a/fs/open.c b/fs/open.c
> index e53af13b5835..f76e960d10ea 100644
> --- a/fs/open.c
> +++ b/fs/open.c
> @@ -852,8 +852,17 @@ static int do_dentry_open(struct file *f,
>  	 * XXX: Huge page cache doesn't support writing yet. Drop all page
>  	 * cache for this file before processing writes.
>  	 */
> -	if ((f->f_mode & FMODE_WRITE) && filemap_nr_thps(inode->i_mapping))
> -		truncate_pagecache(inode, 0);
> +	if (f->f_mode & FMODE_WRITE) {
> +		/*
> +		 * Paired with smp_mb() in collapse_file() to ensure nr_thps
> +		 * is up to date and the update to i_writecount by
> +		 * get_write_access() is visible. Ensures subsequent insertion
> +		 * of THPs into the page cache will fail.
> +		 */
> +		smp_mb();
> +		if (filemap_nr_thps(inode->i_mapping))
> +			truncate_pagecache(inode, 0);
> +	}
>  
>  	return 0;
>  
> diff --git a/mm/khugepaged.c b/mm/khugepaged.c
> index a7d6cb912b05..4c7cc877d5e3 100644
> --- a/mm/khugepaged.c
> +++ b/mm/khugepaged.c
> @@ -459,7 +459,8 @@ static bool hugepage_vma_check(struct vm_area_struct *vma,
>  
>  	/* Read-only file mappings need to be aligned for THP to work. */
>  	if (IS_ENABLED(CONFIG_READ_ONLY_THP_FOR_FS) && vma->vm_file &&
> -	    (vm_flags & VM_DENYWRITE)) {
> +	    !inode_is_open_for_write(vma->vm_file->f_inode) &&
> +	    (vm_flags & VM_EXEC)) {
>  		return IS_ALIGNED((vma->vm_start >> PAGE_SHIFT) - vma->vm_pgoff,
>  				HPAGE_PMD_NR);
>  	}
> @@ -1872,6 +1873,19 @@ static void collapse_file(struct mm_struct *mm,
>  	else {
>  		__mod_lruvec_page_state(new_page, NR_FILE_THPS, nr);
>  		filemap_nr_thps_inc(mapping);
> +		/*
> +		 * Paired with smp_mb() in do_dentry_open() to ensure
> +		 * i_writecount is up to date and the update to nr_thps is
> +		 * visible. Ensures the page cache will be truncated if the
> +		 * file is opened writable.
> +		 */
> +		smp_mb();
> +		if (inode_is_open_for_write(mapping->host)) {
> +			result = SCAN_FAIL;
> +			__mod_lruvec_page_state(new_page, NR_FILE_THPS, -nr);
> +			filemap_nr_thps_dec(mapping);
> +			goto xa_locked;
> +		}
>  	}
>  
>  	if (nr_none) {
> -- 
> 2.31.0.208.g409f899ff0-goog