Re: [PATCH v4 05/21] mm: switch the rmap lock held option off in compat layer
"Lorenzo Stoakes (Oracle)" <[email protected]> Tue, 24 Mar 2026 16:35:10 +0000
| Newsgroups | org.kernel.vger.target-devel,dev.linux.lists.linux-staging,org.infradead.lists.linux-arm-kernel,org.infradead.lists.linux-mtd,org.kernel.vger.linux-doc,org.kernel.vger.linux-fsdevel,org.kernel.vger.linux-hyperv,org.kernel.vger.linux-kernel,org.kernel.vger.linux-scsi,org.kvack.linux-mm |
|---|---|
| Message-ID | <[email protected]> |
On Tue, Mar 24, 2026 at 03:26:28PM +0100, Vlastimil Babka (SUSE) wrote: > On 3/20/26 23:39, Lorenzo Stoakes (Oracle) wrote: > > In the mmap_prepare compatibility layer, we don't need to hold the rmap > > lock, as we are being called from an .mmap handler. > > > > The .mmap_prepare hook, when invoked in the VMA logic, is called prior to > > the VMA being instantiated, but the completion hook is called after the VMA > > is linked into the maple tree, meaning rmap walkers can reach it. > > > > The mmap hook does not link the VMA into the tree, so this cannot happen. > > > > Therefore it's safe to simply disable this in the mmap_prepare > > compatibility layer. > > > > Also update VMA tests code to reflect current compatibility layer state. > > > > Signed-off-by: Lorenzo Stoakes (Oracle) <[email protected]> > > Acked-by: Vlastimil Babka (SUSE) <[email protected]> > > a typo fix below, Andrew can fix locally? > > > --- > > mm/util.c | 6 ++++- > > tools/testing/vma/include/dup.h | 42 +++++++++++++++++---------------- > > 2 files changed, 27 insertions(+), 21 deletions(-) > > > > diff --git a/mm/util.c b/mm/util.c > > index a2cfa0d77c35..182f0f5cc400 100644 > > --- a/mm/util.c > > +++ b/mm/util.c > > @@ -1204,6 +1204,7 @@ int compat_vma_mmap(struct file *file, struct vm_area_struct *vma) > > > > .action.type = MMAP_NOTHING, /* Default */ > > }; > > + struct mmap_action *action = &desc.action; > > int err; > > > > err = vfs_mmap_prepare(file, &desc); > > @@ -1214,8 +1215,11 @@ int compat_vma_mmap(struct file *file, struct vm_area_struct *vma) > > if (err) > > return err; > > > > + /* being invoked from .mmmap means we don't have to enforce this. */ > > .mmap mmmmm map! ;) Andrew - could you fixup in place? Thanks. > > > + action->hide_from_rmap_until_complete = false; > > + > > set_vma_from_desc(vma, &desc); > > - err = mmap_action_complete(vma, &desc.action); > > + err = mmap_action_complete(vma, action); > > if (err) { > > const size_t len = vma_pages(vma) << PAGE_SHIFT; > >