On Tue, Mar 19, 2024 at 01:49:17PM -0700, Yosry Ahmed wrote:
> 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@xxxxxxxxxx> 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().
>
Okay so I had a moment to read it more closely. If we're mapping > 1
pages, and one of the latter pages fails. Then we could leave some
mappings installed.
@Andrew, I think this makes sense. Would you please be able to pick this
up? I'm not running a tree this window. I will try to send out the percpu
hotplug changes I've been forward porting for a while now and try to get
that in a branch for-6.10.
Acked-by: Dennis Zhou <dennis@xxxxxxxxxx>
> I think it is needed to clean up any leftover partial mappings. I am
> not sure about the kmsan example though.
>
Yeah the kmsan example seems like it could be wrong for the same reason,
but I haven't inspected that more closely.
Thanks,
Dennis
> Adding vmalloc reviewers here as well here.
> >
> >
> > > > ---
> > > >
> > > > 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
> > > >
