[PATCH v3 08/14] arm64: exec: Adjust affinity for compat tasks with mismatched 32-bit EL0

Will Deacon will at kernel.org
Thu Nov 19 11:51:51 EST 2020


On Thu, Nov 19, 2020 at 05:44:11PM +0100, Peter Zijlstra wrote:
> On Thu, Nov 19, 2020 at 04:30:36PM +0000, Will Deacon wrote:
> > On Thu, Nov 19, 2020 at 05:19:56PM +0100, Peter Zijlstra wrote:
> > > On Thu, Nov 19, 2020 at 11:06:04AM +0000, Will Deacon wrote:
> > > > On Thu, Nov 19, 2020 at 09:24:07AM +0000, Quentin Perret wrote:
> > > > > On Friday 13 Nov 2020 at 09:37:13 (+0000), Will Deacon wrote:
> > > > > > When exec'ing a 32-bit task on a system with mismatched support for
> > > > > > 32-bit EL0, try to ensure that it starts life on a CPU that can actually
> > > > > > run it.
> > > > > > 
> > > > > > Signed-off-by: Will Deacon <will at kernel.org>
> > > > > > ---
> > > > > >  arch/arm64/kernel/process.c | 12 +++++++++++-
> > > > > >  1 file changed, 11 insertions(+), 1 deletion(-)
> > > > > > 
> > > > > > diff --git a/arch/arm64/kernel/process.c b/arch/arm64/kernel/process.c
> > > > > > index 1540ab0fbf23..17b94007fed4 100644
> > > > > > --- a/arch/arm64/kernel/process.c
> > > > > > +++ b/arch/arm64/kernel/process.c
> > > > > > @@ -625,6 +625,16 @@ unsigned long arch_align_stack(unsigned long sp)
> > > > > >  	return sp & ~0xf;
> > > > > >  }
> > > > > >  
> > > > > > +static void adjust_compat_task_affinity(struct task_struct *p)
> > > > > > +{
> > > > > > +	const struct cpumask *mask = system_32bit_el0_cpumask();
> > > > > > +
> > > > > > +	if (restrict_cpus_allowed_ptr(p, mask))
> > > > > > +		set_cpus_allowed_ptr(p, mask);
> > > > > 
> > > > > My understanding of this call to set_cpus_allowed_ptr() is that you're
> > > > > mimicking the hotplug vs affinity case behaviour in some ways. That is,
> > > > > if a task is pinned to a CPU and userspace hotplugs that CPU, then the
> > > > > kernel will reset the affinity of the task to the remaining online CPUs.
> > > > 
> > > > Correct. It looks to the 32-bit application like all the 64-bit-only CPUs
> > > > were hotplugged off at the point of the execve().
> > > 
> > > This doesn't respect cpusets though :/
> > 
> > How does that differ from select_fallback_rq()?
> 
> That has that cpuset state it tries first, only if that fails does it do
> the possible mask.

Fair enough. Lemme try something similar here before overriding the mask
entirely...

Will



More information about the linux-arm-kernel mailing list