by Lorenzo Stoakes

[permalink] [raw]

Subject: Re: [PATCH v2 4/5] mm: abstract merge for new VMAs into vma_merge_new_vma()

On Tue, Oct 10, 2023 at 09:51:40PM -0400, Liam R. Howlett wrote:
> * Lorenzo Stoakes <[email protected]> [231009 16:53]:
> > Only in mmap_region() and copy_vma() do we attempt to merge VMAs which
> > occupy entirely new regions of virtual memory.
> >
> > We can abstract this logic and make the intent of this invocations of it
> > completely explicit, rather than invoking vma_merge() with an inscrutable
> > wall of parameters.
> >
> > This also paves the way for a simplification of the core vma_merge()
> > implementation, as we seek to make it entirely an implementation detail.
> >
> > Note that on mmap_region(), VMA fields are initialised to zero, so we can
> > simply reference these rather than explicitly specifying NULL.
>
> I don't think that's accurate.. mmap_region() sets the start, end,
> offset, flags. It also passes this vma into a driver, so I'm not sure
> we can rely on them being anything after that? The whole reason
> vma_merge() is attempted in this case is because the driver may have
> changed vma->vm_flags on us. Your way may actually be better since the
> driver may set something we assume is NULL today.

Yeah I think I wasn't clear here - I meant to say that we memset -> 0 so
all fields that are not specified (e.g. not start, end, offset, flags).

However you make a very good point re: the driver, which I hadn't thought
of, also it's worth saying here that we specifically only do this for a
file-backed mapping just for complete clarity.

I will add a note to this part of the v3 series asking Andrew to update the
comment.

>
> >
> > Reviewed-by: Vlastimil Babka <[email protected]>
> > Signed-off-by: Lorenzo Stoakes <[email protected]>
> > ---
> > mm/mmap.c | 27 ++++++++++++++++++++-------
> > 1 file changed, 20 insertions(+), 7 deletions(-)
> >
> > diff --git a/mm/mmap.c b/mm/mmap.c
> > index 17c0dcfb1527..33aafd23823b 100644
> > --- a/mm/mmap.c
> > +++ b/mm/mmap.c
> > @@ -2482,6 +2482,22 @@ struct vm_area_struct *vma_modify(struct vma_iterator *vmi,
> > return NULL;
> > }
> >
> > +/*
> > + * Attempt to merge a newly mapped VMA with those adjacent to it. The caller
> > + * must ensure that [start, end) does not overlap any existing VMA.
> > + */
> > +static struct vm_area_struct *vma_merge_new_vma(struct vma_iterator *vmi,
> > + struct vm_area_struct *prev,
> > + struct vm_area_struct *vma,
> > + unsigned long start,
> > + unsigned long end,
> > + pgoff_t pgoff)
>
> It's not a coding style, but if you used two tabs here, it may make this
> more condensed.

Checkpatch shouts at me about aligning to the paren, I obviously could just
put "static struct vm_area_struct *" on the line before to make this a bit
better though. If we go to a v4 will fix, otherwise I think probably ok to
leave even if a bit squished for now?

>
> > +{
> > + return vma_merge(vmi, vma->vm_mm, prev, start, end, vma->vm_flags,
> > + vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
> > + vma->vm_userfaultfd_ctx, anon_vma_name(vma));
> > +}
> > +
> > /*
> > * do_vmi_align_munmap() - munmap the aligned region from @start to @end.
> > * @vmi: The vma iterator
> > @@ -2837,10 +2853,9 @@ unsigned long mmap_region(struct file *file, unsigned long addr,
> > * vma again as we may succeed this time.
> > */
> > if (unlikely(vm_flags != vma->vm_flags && prev)) {
> > - merge = vma_merge(&vmi, mm, prev, vma->vm_start,
> > - vma->vm_end, vma->vm_flags, NULL,
> > - vma->vm_file, vma->vm_pgoff, NULL,
> > - NULL_VM_UFFD_CTX, NULL);
> > + merge = vma_merge_new_vma(&vmi, prev, vma,
> > + vma->vm_start, vma->vm_end,
> > + pgoff);
> └ vma->vm_pgoff
> > if (merge) {
> > /*
> > * ->mmap() can change vma->vm_file and fput
> > @@ -3382,9 +3397,7 @@ struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
> > if (new_vma && new_vma->vm_start < addr + len)
> > return NULL; /* should never get here */
> >
> > - new_vma = vma_merge(&vmi, mm, prev, addr, addr + len, vma->vm_flags,
> > - vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
> > - vma->vm_userfaultfd_ctx, anon_vma_name(vma));
> > + new_vma = vma_merge_new_vma(&vmi, prev, vma, addr, addr + len, pgoff);
> > if (new_vma) {
> > /*
> > * Source vma may have been merged into new_vma
> > --
> > 2.42.0
> >