Re: [PATCH] dynamically allocate struct cpumask

[Nick Desaulniers <nick.desaulniers@xxxxxxxxx>]

On Mon, Jan 27, 2020 at 12:09 AM Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> On Sun, Jan 26, 2020 at 11:16:02PM -0800, Nick Desaulniers wrote:
> > This helps avoid avoid a potentially large stack allocation.
> >
> > When building with:
> > $ make CC=clang arch/x86/ CFLAGS=-Wframe-larger-than=1000
> > The following warning is observed:
> > arch/x86/kernel/kvm.c:494:13: warning: stack frame size of 1064 bytes in
> > function 'kvm_send_ipi_mask_allbutself' [-Wframe-larger-than=]
> > static void kvm_send_ipi_mask_allbutself(const struct cpumask *mask, int
> > vector)
> >             ^
> > Debugging with:
> > https://github.com/ClangBuiltLinux/frame-larger-than
> > via:
> > $ python3 frame_larger_than.py arch/x86/kernel/kvm.o \
> >   kvm_send_ipi_mask_allbutself
> > points to the stack allocated `struct cpumask newmask` in
> > `kvm_send_ipi_mask_allbutself`. The size of a `struct cpumask` is
> > potentially large, as it's CONFIG_NR_CPUS divided by BITS_PER_LONG for
> > the target architecture. CONFIG_NR_CPUS for X86_64 can be as high as
> > 8192, making a single instance of a `struct cpumask` 1024 B.
> >
> > Signed-off-by: Nick Desaulniers <nick.desaulniers@xxxxxxxxx>
> > ---
> >  arch/x86/kernel/kvm.c | 10 ++++++----
> >  1 file changed, 6 insertions(+), 4 deletions(-)
> >
> > diff --git a/arch/x86/kernel/kvm.c b/arch/x86/kernel/kvm.c
> > index 32ef1ee733b7..d41c0a0d62a2 100644
> > --- a/arch/x86/kernel/kvm.c
> > +++ b/arch/x86/kernel/kvm.c
> > @@ -494,13 +494,15 @@ static void kvm_send_ipi_mask(const struct cpumask *mask, int vector)
> >  static void kvm_send_ipi_mask_allbutself(const struct cpumask *mask, int vector)
> >  {
> >       unsigned int this_cpu = smp_processor_id();
> > -     struct cpumask new_mask;
>
> Right, on stack cpumask is definitely dodgy.
>
> > +     struct cpumask *new_mask;
> >       const struct cpumask *local_mask;
> >
> > -     cpumask_copy(&new_mask, mask);
> > -     cpumask_clear_cpu(this_cpu, &new_mask);
> > -     local_mask = &new_mask;
> > +     new_mask = kmalloc(sizeof(*new_mask), GFP_KERNEL);

Probably should check for allocation failure, d'oh!

> > +     cpumask_copy(new_mask, mask);
> > +     cpumask_clear_cpu(this_cpu, new_mask);
> > +     local_mask = new_mask;
> >       __send_ipi_mask(local_mask, vector);
> > +     kfree(new_mask);
> >  }
>
> One alternative approach is adding the inverse of cpu_bit_bitmap. I'm
> not entirely sure how often we need the all-but-self mask, but ISTR
> there were other places too.