Re: [PATCH v3 5/5] mm: defer second attempt at merge on mmap()

[Vlastimil Babka]

On 10/25/24 14:26, Lorenzo Stoakes wrote:
> Rather than trying to merge again when ostensibly allocating a new VMA,
> instead defer until the VMA is added and attempt to merge the existing
> range.
> 
> This way we have no complicated unwinding logic midway through the process
> of mapping the VMA.
> 
> In addition this removes limitations on the VMA not being able to be the
> first in the virtual memory address space which was previously implicitly
> required.
> 
> In theory, for this very same reason, we should unconditionally attempt
> merge here, however this is likely to have a performance impact so it is
> better to avoid this given the unlikely outcome of a merge.
> 
> Reviewed-by: Vlastimil Babka <vbabka@xxxxxxx>
> Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@xxxxxxxxxx>
> ---
>  mm/vma.c | 55 ++++++++++++++-----------------------------------------
>  1 file changed, 14 insertions(+), 41 deletions(-)
> 
> diff --git a/mm/vma.c b/mm/vma.c
> index 7c690be67910..7194f9449c60 100644
> --- a/mm/vma.c
> +++ b/mm/vma.c
> @@ -19,6 +19,7 @@ struct mmap_state {
>  	struct file *file;
> 
>  	unsigned long charged;
> +	bool retry_merge;
> 
>  	struct vm_area_struct *prev;
>  	struct vm_area_struct *next;
> @@ -2278,8 +2279,9 @@ static int __mmap_prepare(struct mmap_state *map, struct list_head *uf)
>  	return 0;
>  }
> 
> +
>  static int __mmap_new_file_vma(struct mmap_state *map,
> -			       struct vm_area_struct **vmap, bool *mergedp)
> +			       struct vm_area_struct **vmap)

AFAICS **vmap could become *vma now

>  {
>  	struct vma_iterator *vmi = map->vmi;
>  	struct vm_area_struct *vma = *vmap;