Re: [PATCH] arm/arm64: xen: Fix to convert percpu address to gfn correctly

[Stefano Stabellini]

On Mon, 5 Oct 2020, Julien Grall wrote:
> Hi Masami,
> 
> On 05/10/2020 14:39, Masami Hiramatsu wrote:
> > Use per_cpu_ptr_to_phys() instead of virt_to_phys() for per-cpu
> > address conversion.
> > 
> > In xen_starting_cpu(), per-cpu xen_vcpu_info address is converted
> > to gfn by virt_to_gfn() macro. However, since the virt_to_gfn(v)
> > assumes the given virtual address is in contiguous kernel memory
> > area, it can not convert the per-cpu memory if it is allocated on
> > vmalloc area (depends on CONFIG_SMP).
> 
> Are you sure about this? I have a .config with CONFIG_SMP=y where the per-cpu
> region for CPU0 is allocated outside of vmalloc area.
> 
> However, I was able to trigger the bug as soon as CONFIG_NUMA_BALANCING was
> enabled.

I cannot reproduce the issue with defconfig, but I can with Masami's
kconfig.

If I disable just CONFIG_NUMA_BALANCING from Masami's kconfig, the
problem still appears.

If I disable CONFIG_NUMA from Masami's kconfig, it works, which is
strange because CONFIG_NUMA is enabled in defconfig, and defconfig
works.


> [...]
> 
> > Fixes: 250c9af3d831 ("arm/xen: Add support for 64KB page granularity")
> 
> FWIW, I think the bug was already present before 250c9af3d831.

Yeah, I bet 250c9af3d831 is not what introduced the issue. Whatever
caused virt_to_phys to stop working on vmalloc'ed addresses is the cause
of the problem. It is something that went in 5.9 (5.8 works) but I don't
know what for sure.


> > Signed-off-by: Masami Hiramatsu <mhiramat@xxxxxxxxxx>
> > ---
> >   arch/arm/xen/enlighten.c |    2 +-
> >   include/xen/arm/page.h   |    3 +++
> >   2 files changed, 4 insertions(+), 1 deletion(-)
> > 
> > diff --git a/arch/arm/xen/enlighten.c b/arch/arm/xen/enlighten.c
> > index e93145d72c26..a6ab3689b2f4 100644
> > --- a/arch/arm/xen/enlighten.c
> > +++ b/arch/arm/xen/enlighten.c
> > @@ -150,7 +150,7 @@ static int xen_starting_cpu(unsigned int cpu)
> >   	pr_info("Xen: initializing cpu%d\n", cpu);
> >   	vcpup = per_cpu_ptr(xen_vcpu_info, cpu);
> >   -	info.mfn = virt_to_gfn(vcpup);
> > +	info.mfn = percpu_to_gfn(vcpup);
> >   	info.offset = xen_offset_in_page(vcpup);
> >     	err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, xen_vcpu_nr(cpu),
> > diff --git a/include/xen/arm/page.h b/include/xen/arm/page.h
> > index 39df751d0dc4..ac1b65470563 100644
> > --- a/include/xen/arm/page.h
> > +++ b/include/xen/arm/page.h
> > @@ -83,6 +83,9 @@ static inline unsigned long bfn_to_pfn(unsigned long bfn)
> >   	})
> >   #define gfn_to_virt(m)		(__va(gfn_to_pfn(m) <<
> > XEN_PAGE_SHIFT))
> >   +#define percpu_to_gfn(v)	\
> > +	(pfn_to_gfn(per_cpu_ptr_to_phys(v) >> XEN_PAGE_SHIFT))
> > +
> >   /* Only used in PV code. But ARM guests are always HVM. */
> >   static inline xmaddr_t arbitrary_virt_to_machine(void *vaddr)
> >   {


The fix is fine for me. I tested it and it works. We need to remove the
"Fixes:" line from the commit message. Ideally, replacing it with a
reference to what is the source of the problem.

Aside from that:

Reviewed-by: Stefano Stabellini <sstabellini@xxxxxxxxxx>