Re: [PATCH] percpu: clean up all mappings when pcpu_map_pages() fails

[Yosry Ahmed]

On Tue, Mar 19, 2024 at 1:32 PM Dennis Zhou <dennis@xxxxxxxxxx> wrote:
>
> Hi Yosry,
>
> On Tue, Mar 19, 2024 at 01:08:26PM -0700, Yosry Ahmed wrote:
> > On Mon, Mar 11, 2024 at 12:43 PM Yosry Ahmed <yosryahmed@googlecom> wrote:
> > >
> > > In pcpu_map_pages(), if __pcpu_map_pages() fails on a CPU, we call
> > > __pcpu_unmap_pages() to clean up mappings on all CPUs where mappings
> > > were created, but not on the CPU where __pcpu_map_pages() fails.
> > >
> > > __pcpu_map_pages() and __pcpu_unmap_pages() are wrappers around
> > > vmap_pages_range_noflush() and vunmap_range_noflush(). All other callers
> > > of vmap_pages_range_noflush() call vunmap_range_noflush() when mapping
> > > fails, except pcpu_map_pages(). The reason could be that partial
> > > mappings may be left behind from a failed mapping attempt.
> > >
> > > Call __pcpu_unmap_pages() for the failed CPU as well in
> > > pcpu_map_pages().
> > >
> > > This was found by code inspection, no failures or bugs were observed.
> > >
> > > Signed-off-by: Yosry Ahmed <yosryahmed@xxxxxxxxxx>
> >
> > Any thoughts about this change? Should I resend next week after the
> > merge window?
> >
>
> Sorry for the delay.
>
> I'm looking at the code from mm/kmsan/hooks.c kmsan_ioremap_page_range().
> It seems like __vunmap_range_noflush() is called on error for
> successfully mapped pages similar to how it's being done in percpu-vm.c.

You  picked an unconventional example to compare against :)

>
> I haven't read in depth the expectations of vmap_pages_range_noflush()
> but on first glance it doesn't seem like percpu is operating out of the
> ordinary?

I was looking at vm_map_ram(), vmap(), and __vmalloc_area_node(). They
all call vmap_pages_range()-> vmap_pages_range_noflush().

When vmap_pages_range() fails:
- vm_map_ram() calls
vm_unmap_ram()->free_unmap_vmap_area()->vunmap_range_noflush().
- vmap() calls vunmap()->remove_vm_area()->free_unmap_vmap_area()->
vunmap_range_noflush().
- __vmalloc_area_node() calls
vfree()->remove_vm_area()->free_unmap_vmap_area()->
vunmap_range_noflush().

I think it is needed to clean up any leftover partial mappings. I am
not sure about the kmsan example though.

Adding vmalloc reviewers here as well here.
>
> Thanks,
> Dennis
>
> > > ---
> > >
> > > Perhaps the reason __pcpu_unmap_pages() is not currently being called
> > > for the failed CPU is that the size and alignment requirements make sure
> > > we never leave any partial mappings behind? I have no idea. Nonetheless,
> > > I think we want this change as that could be fragile, and is
> > > inconsistent with other callers.
> > >
> > > ---
> > >  mm/percpu-vm.c | 4 ++--
> > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/mm/percpu-vm.c b/mm/percpu-vm.c
> > > index 2054c9213c433..cd69caf6aa8d8 100644
> > > --- a/mm/percpu-vm.c
> > > +++ b/mm/percpu-vm.c
> > > @@ -231,10 +231,10 @@ static int pcpu_map_pages(struct pcpu_chunk *chunk,
> > >         return 0;
> > >  err:
> > >         for_each_possible_cpu(tcpu) {
> > > -               if (tcpu == cpu)
> > > -                       break;
> > >                 __pcpu_unmap_pages(pcpu_chunk_addr(chunk, tcpu, page_start),
> > >                                    page_end - page_start);
> > > +               if (tcpu == cpu)
> > > +                       break;
> > >         }
> > >         pcpu_post_unmap_tlb_flush(chunk, page_start, page_end);
> > >         return err;
> > > --
> > > 2.44.0.278.ge034bb2e1d-goog
> > >